Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 05:08:35AM -0500, Pankaj Gupta wrote: > > > > > On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: > > > > > > > > > On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin > > > wrote: > > > >On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: > > > >> On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang > > > >>wrote: > > > >> > > > > >> > > > > >> >On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin > > > >>wrote: > > > >> >>On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > > > >> >>> Hello: > > > >> >>> We used to orphan packets before transmission for virtio-net. This > > > >> >>>breaks > > > >> >>> socket accounting and can lead serveral functions won't work, e.g: > > > >> >>> - Byte Queue Limit depends on tx completion nofication to work. > > > >> >>> - Packet Generator depends on tx completion nofication for the last > > > >> >>> transmitted packet to complete. > > > >> >>> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > > > >> >>>work. > > > >> >>> This series tries to solve the issue by enabling tx interrupts. To > > > >> >>>minize > > > >> >>> the performance impacts of this, several optimizations were used: > > > >> >>> - In guest side, virtqueue_enable_cb_delayed() was used to delay > > > >>the > > > >> >>>tx > > > >> >>> interrupt untile 3/4 pending packets were sent. > > > >> >>> - In host side, interrupt coalescing were used to reduce tx > > > >> >>>interrupts. > > > >> >>> Performance test results[1] (tx-frames 16 tx-usecs 16) shows: > > > >> >>> - For guest receiving. No obvious regression on throughput were > > > >> >>> noticed. More cpu utilization were noticed in few cases. > > > >> >>> - For guest transmission. Very huge improvement on througput for > > > >> >>>small > > > >> >>> packet transmission were noticed. This is expected since TSQ and > > > >> >>>other > > > >> >>> optimization for small packet transmission work after tx > > > >>interrupt. > > > >> >>>But > > > >> >>> will use more cpu for large packets. > > > >> >>> - For TCP_RR, regression (10% on transaction rate and cpu > > > >> >>>utilization) were > > > >> >>> found. Tx interrupt won't help but cause overhead in this case. > > > >> >>>Using > > > >> >>> more aggressive coalescing parameters may help to reduce the > > > >> >>>regression. > > > >> >> > > > >> >>OK, you do have posted coalescing patches - does it help any? > > > >> > > > > >> >Helps a lot. > > > >> > > > > >> >For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) > > > >> >For small packet TX, it increases 33% - 245% throughput. (reduce about > > > >>60% > > > >> >inters) > > > >> >For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx > > > >>intrs) > > > >> > > > > >> >> > > > >> >>I'm not sure the regression is due to interrupts. > > > >> >>It would make sense for CPU but why would it > > > >> >>hurt transaction rate? > > > >> > > > > >> >Anyway guest need to take some cycles to handle tx interrupts. > > > >> >And transaction rate does increase if we coalesces more tx interurpts. > > > >> >> > > > >> >> > > > >> >>It's possible that we are deferring kicks too much due to BQL. > > > >> >> > > > >> >>As an experiment: do we get any of it back if we do > > > >> >>-if (kick || netif_xmit_stopped(txq)) > > > >> >>-virtqueue_kick(sq->vq); > > > >> >>+virtqueue_kick(sq->vq); > > > >> >>? > > > >> > > > > >> > > > > >> >I will try, but during TCP_RR, at most 1 packets were pending, > > > >> >I suspect if BQL can help in this case. > > > >> Looks like this helps a lot in multiple sessions of TCP_RR. > > > > > > > >so what's faster > > > > BQL + kick each packet > > > > no BQL > > > >? > > > > > > Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious > > > differences. > > > > > > May need a complete benchmark to see. > > > > Okay so going forward something like BQL + kick each packet > > might be a good solution. > > The advantage of BQL is that it works without GSO. > > For example, now that we don't do UFO, you might > > see significant gains with UDP. > > If I understand correctly, it can also help for small packet > regr. in multiqueue scenario? Well BQL generally should only be active for 1:1 mappings. > Would be nice to see the perf. numbers > with multi-queue for small packets streams. > > > > > > > > > > > > > > > >> How about move the BQL patch out of this series? > > > >> Let's first converge tx interrupt and then introduce it? > > > >> (e.g with kicking after queuing X bytes?) > > > > > > > >Sounds good. > > -- > > To unsubscribe from this list: send the line "unsubscribe netdev" in > > the body of a message to majord...@vger.kernel.org > > More majordomo info at http://vger.kernel.org/majordomo-info.html > > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
> > On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: > > > > > > On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin wrote: > > >On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: > > >> On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang > > >>wrote: > > >> > > > >> > > > >> >On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin > > >>wrote: > > >> >>On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > > >> >>> Hello: > > >> >>> We used to orphan packets before transmission for virtio-net. This > > >> >>>breaks > > >> >>> socket accounting and can lead serveral functions won't work, e.g: > > >> >>> - Byte Queue Limit depends on tx completion nofication to work. > > >> >>> - Packet Generator depends on tx completion nofication for the last > > >> >>> transmitted packet to complete. > > >> >>> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > > >> >>>work. > > >> >>> This series tries to solve the issue by enabling tx interrupts. To > > >> >>>minize > > >> >>> the performance impacts of this, several optimizations were used: > > >> >>> - In guest side, virtqueue_enable_cb_delayed() was used to delay > > >>the > > >> >>>tx > > >> >>> interrupt untile 3/4 pending packets were sent. > > >> >>> - In host side, interrupt coalescing were used to reduce tx > > >> >>>interrupts. > > >> >>> Performance test results[1] (tx-frames 16 tx-usecs 16) shows: > > >> >>> - For guest receiving. No obvious regression on throughput were > > >> >>> noticed. More cpu utilization were noticed in few cases. > > >> >>> - For guest transmission. Very huge improvement on througput for > > >> >>>small > > >> >>> packet transmission were noticed. This is expected since TSQ and > > >> >>>other > > >> >>> optimization for small packet transmission work after tx > > >>interrupt. > > >> >>>But > > >> >>> will use more cpu for large packets. > > >> >>> - For TCP_RR, regression (10% on transaction rate and cpu > > >> >>>utilization) were > > >> >>> found. Tx interrupt won't help but cause overhead in this case. > > >> >>>Using > > >> >>> more aggressive coalescing parameters may help to reduce the > > >> >>>regression. > > >> >> > > >> >>OK, you do have posted coalescing patches - does it help any? > > >> > > > >> >Helps a lot. > > >> > > > >> >For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) > > >> >For small packet TX, it increases 33% - 245% throughput. (reduce about > > >>60% > > >> >inters) > > >> >For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx > > >>intrs) > > >> > > > >> >> > > >> >>I'm not sure the regression is due to interrupts. > > >> >>It would make sense for CPU but why would it > > >> >>hurt transaction rate? > > >> > > > >> >Anyway guest need to take some cycles to handle tx interrupts. > > >> >And transaction rate does increase if we coalesces more tx interurpts. > > >> >> > > >> >> > > >> >>It's possible that we are deferring kicks too much due to BQL. > > >> >> > > >> >>As an experiment: do we get any of it back if we do > > >> >>-if (kick || netif_xmit_stopped(txq)) > > >> >>-virtqueue_kick(sq->vq); > > >> >>+virtqueue_kick(sq->vq); > > >> >>? > > >> > > > >> > > > >> >I will try, but during TCP_RR, at most 1 packets were pending, > > >> >I suspect if BQL can help in this case. > > >> Looks like this helps a lot in multiple sessions of TCP_RR. > > > > > >so what's faster > > > BQL + kick each packet > > > no BQL > > >? > > > > Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious > > differences. > > > > May need a complete benchmark to see. > > Okay so going forward something like BQL + kick each packet > might be a good solution. > The advantage of BQL is that it works without GSO. > For example, now that we don't do UFO, you might > see significant gains with UDP. If I understand correctly, it can also help for small packet regr. in multiqueue scenario? Would be nice to see the perf. numbers with multi-queue for small packets streams. > > > > > > > > > > >> How about move the BQL patch out of this series? > > >> Let's first converge tx interrupt and then introduce it? > > >> (e.g with kicking after queuing X bytes?) > > > > > >Sounds good. > -- > To unsubscribe from this list: send the line "unsubscribe netdev" in > the body of a message to majord...@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 10:00:06AM +, David Laight wrote: > From: Jason Wang > > > On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > > >> Hello: > > >> > > >> We used to orphan packets before transmission for virtio-net. This > > >> breaks > > >> socket accounting and can lead serveral functions won't work, e.g: > > >> > > >> - Byte Queue Limit depends on tx completion nofication to work. > > >> - Packet Generator depends on tx completion nofication for the last > > >>transmitted packet to complete. > > >> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > > >> work. > > >> > > >> This series tries to solve the issue by enabling tx interrupts. To > > >> minize > > >> the performance impacts of this, several optimizations were used: > > >> > > >> - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx > > >>interrupt untile 3/4 pending packets were sent. > > Doesn't that give problems for intermittent transmits? > > ... > > David > No because it has not effect in that case. -- MST -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
RE: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
From: Jason Wang > > On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > >> Hello: > >> > >> We used to orphan packets before transmission for virtio-net. This > >> breaks > >> socket accounting and can lead serveral functions won't work, e.g: > >> > >> - Byte Queue Limit depends on tx completion nofication to work. > >> - Packet Generator depends on tx completion nofication for the last > >>transmitted packet to complete. > >> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > >> work. > >> > >> This series tries to solve the issue by enabling tx interrupts. To > >> minize > >> the performance impacts of this, several optimizations were used: > >> > >> - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx > >>interrupt untile 3/4 pending packets were sent. Doesn't that give problems for intermittent transmits? ... David
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: > > > On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin wrote: > >On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: > >> On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang > >>wrote: > >> > > >> > > >> >On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin > >>wrote: > >> >>On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > >> >>> Hello: > >> >>> We used to orphan packets before transmission for virtio-net. This > >> >>>breaks > >> >>> socket accounting and can lead serveral functions won't work, e.g: > >> >>> - Byte Queue Limit depends on tx completion nofication to work. > >> >>> - Packet Generator depends on tx completion nofication for the last > >> >>> transmitted packet to complete. > >> >>> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > >> >>>work. > >> >>> This series tries to solve the issue by enabling tx interrupts. To > >> >>>minize > >> >>> the performance impacts of this, several optimizations were used: > >> >>> - In guest side, virtqueue_enable_cb_delayed() was used to delay > >>the > >> >>>tx > >> >>> interrupt untile 3/4 pending packets were sent. > >> >>> - In host side, interrupt coalescing were used to reduce tx > >> >>>interrupts. > >> >>> Performance test results[1] (tx-frames 16 tx-usecs 16) shows: > >> >>> - For guest receiving. No obvious regression on throughput were > >> >>> noticed. More cpu utilization were noticed in few cases. > >> >>> - For guest transmission. Very huge improvement on througput for > >> >>>small > >> >>> packet transmission were noticed. This is expected since TSQ and > >> >>>other > >> >>> optimization for small packet transmission work after tx > >>interrupt. > >> >>>But > >> >>> will use more cpu for large packets. > >> >>> - For TCP_RR, regression (10% on transaction rate and cpu > >> >>>utilization) were > >> >>> found. Tx interrupt won't help but cause overhead in this case. > >> >>>Using > >> >>> more aggressive coalescing parameters may help to reduce the > >> >>>regression. > >> >> > >> >>OK, you do have posted coalescing patches - does it help any? > >> > > >> >Helps a lot. > >> > > >> >For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) > >> >For small packet TX, it increases 33% - 245% throughput. (reduce about > >>60% > >> >inters) > >> >For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx > >>intrs) > >> > > >> >> > >> >>I'm not sure the regression is due to interrupts. > >> >>It would make sense for CPU but why would it > >> >>hurt transaction rate? > >> > > >> >Anyway guest need to take some cycles to handle tx interrupts. > >> >And transaction rate does increase if we coalesces more tx interurpts. > >> >> > >> >> > >> >>It's possible that we are deferring kicks too much due to BQL. > >> >> > >> >>As an experiment: do we get any of it back if we do > >> >>-if (kick || netif_xmit_stopped(txq)) > >> >>-virtqueue_kick(sq->vq); > >> >>+virtqueue_kick(sq->vq); > >> >>? > >> > > >> > > >> >I will try, but during TCP_RR, at most 1 packets were pending, > >> >I suspect if BQL can help in this case. > >> Looks like this helps a lot in multiple sessions of TCP_RR. > > > >so what's faster > > BQL + kick each packet > > no BQL > >? > > Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious > differences. > > May need a complete benchmark to see. Okay so going forward something like BQL + kick each packet might be a good solution. The advantage of BQL is that it works without GSO. For example, now that we don't do UFO, you might see significant gains with UDP. > > > > > >> How about move the BQL patch out of this series? > >> Let's first converge tx interrupt and then introduce it? > >> (e.g with kicking after queuing X bytes?) > > > >Sounds good. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin wrote: On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang wrote: > > >On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin wrote: >>On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: >>> Hello: >>> We used to orphan packets before transmission for virtio-net. This >>>breaks >>> socket accounting and can lead serveral functions won't work, e.g: >>> - Byte Queue Limit depends on tx completion nofication to work. >>> - Packet Generator depends on tx completion nofication for the last >>> transmitted packet to complete. >>> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to >>>work. >>> This series tries to solve the issue by enabling tx interrupts. To >>>minize >>> the performance impacts of this, several optimizations were used: >>> - In guest side, virtqueue_enable_cb_delayed() was used to delay the >>>tx >>> interrupt untile 3/4 pending packets were sent. >>> - In host side, interrupt coalescing were used to reduce tx >>>interrupts. >>> Performance test results[1] (tx-frames 16 tx-usecs 16) shows: >>> - For guest receiving. No obvious regression on throughput were >>> noticed. More cpu utilization were noticed in few cases. >>> - For guest transmission. Very huge improvement on througput for >>>small >>> packet transmission were noticed. This is expected since TSQ and >>>other >>> optimization for small packet transmission work after tx interrupt. >>>But >>> will use more cpu for large packets. >>> - For TCP_RR, regression (10% on transaction rate and cpu >>>utilization) were >>> found. Tx interrupt won't help but cause overhead in this case. >>>Using >>> more aggressive coalescing parameters may help to reduce the >>>regression. >> >>OK, you do have posted coalescing patches - does it help any? > >Helps a lot. > >For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) >For small packet TX, it increases 33% - 245% throughput. (reduce about 60% >inters) >For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) > >> >>I'm not sure the regression is due to interrupts. >>It would make sense for CPU but why would it >>hurt transaction rate? > >Anyway guest need to take some cycles to handle tx interrupts. >And transaction rate does increase if we coalesces more tx interurpts. >> >> >>It's possible that we are deferring kicks too much due to BQL. >> >>As an experiment: do we get any of it back if we do >>-if (kick || netif_xmit_stopped(txq)) >>-virtqueue_kick(sq->vq); >>+virtqueue_kick(sq->vq); >>? > > >I will try, but during TCP_RR, at most 1 packets were pending, >I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious differences. May need a complete benchmark to see. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: > > > On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang wrote: > > > > > >On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin wrote: > >>On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > >>> Hello: > >>> We used to orphan packets before transmission for virtio-net. This > >>>breaks > >>> socket accounting and can lead serveral functions won't work, e.g: > >>> - Byte Queue Limit depends on tx completion nofication to work. > >>> - Packet Generator depends on tx completion nofication for the last > >>> transmitted packet to complete. > >>> - TCP Small Queue depends on proper accounting of sk_wmem_alloc to > >>>work. > >>> This series tries to solve the issue by enabling tx interrupts. To > >>>minize > >>> the performance impacts of this, several optimizations were used: > >>> - In guest side, virtqueue_enable_cb_delayed() was used to delay the > >>>tx > >>> interrupt untile 3/4 pending packets were sent. > >>> - In host side, interrupt coalescing were used to reduce tx > >>>interrupts. > >>> Performance test results[1] (tx-frames 16 tx-usecs 16) shows: > >>> - For guest receiving. No obvious regression on throughput were > >>> noticed. More cpu utilization were noticed in few cases. > >>> - For guest transmission. Very huge improvement on througput for > >>>small > >>> packet transmission were noticed. This is expected since TSQ and > >>>other > >>> optimization for small packet transmission work after tx interrupt. > >>>But > >>> will use more cpu for large packets. > >>> - For TCP_RR, regression (10% on transaction rate and cpu > >>>utilization) were > >>> found. Tx interrupt won't help but cause overhead in this case. > >>>Using > >>> more aggressive coalescing parameters may help to reduce the > >>>regression. > >> > >>OK, you do have posted coalescing patches - does it help any? > > > >Helps a lot. > > > >For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) > >For small packet TX, it increases 33% - 245% throughput. (reduce about 60% > >inters) > >For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) > > > >> > >>I'm not sure the regression is due to interrupts. > >>It would make sense for CPU but why would it > >>hurt transaction rate? > > > >Anyway guest need to take some cycles to handle tx interrupts. > >And transaction rate does increase if we coalesces more tx interurpts. > >> > >> > >>It's possible that we are deferring kicks too much due to BQL. > >> > >>As an experiment: do we get any of it back if we do > >>-if (kick || netif_xmit_stopped(txq)) > >>-virtqueue_kick(sq->vq); > >>+virtqueue_kick(sq->vq); > >>? > > > > > >I will try, but during TCP_RR, at most 1 packets were pending, > >I suspect if BQL can help in this case. > > Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? > How about move the BQL patch out of this series? > > Let's first converge tx interrupt and then introduce it? > (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq->vq); +virtqueue_kick(sq->vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. Okay, let me try to see if this help. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. Okay, let me try to see if this help. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin m...@redhat.com wrote: On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious differences. May need a complete benchmark to see. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin m...@redhat.com wrote: On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious differences. May need a complete benchmark to see. Okay so going forward something like BQL + kick each packet might be a good solution. The advantage of BQL is that it works without GSO. For example, now that we don't do UFO, you might see significant gains with UDP. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
RE: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
From: Jason Wang On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. Doesn't that give problems for intermittent transmits? ... David
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 10:00:06AM +, David Laight wrote: From: Jason Wang On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. Doesn't that give problems for intermittent transmits? ... David No because it has not effect in that case. -- MST -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin m...@redhat.com wrote: On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious differences. May need a complete benchmark to see. Okay so going forward something like BQL + kick each packet might be a good solution. The advantage of BQL is that it works without GSO. For example, now that we don't do UFO, you might see significant gains with UDP. If I understand correctly, it can also help for small packet regr. in multiqueue scenario? Would be nice to see the perf. numbers with multi-queue for small packets streams. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Tue, Dec 02, 2014 at 05:08:35AM -0500, Pankaj Gupta wrote: On Tue, Dec 02, 2014 at 09:59:48AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 5:43 PM, Michael S. Tsirkin m...@redhat.com wrote: On Tue, Dec 02, 2014 at 08:15:02AM +0008, Jason Wang wrote: On Tue, Dec 2, 2014 at 11:15 AM, Jason Wang jasow...@redhat.com wrote: On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. Looks like this helps a lot in multiple sessions of TCP_RR. so what's faster BQL + kick each packet no BQL ? Quick and manual tests (TCP_RR 64, TCP_STREAM 512) does not show obvious differences. May need a complete benchmark to see. Okay so going forward something like BQL + kick each packet might be a good solution. The advantage of BQL is that it works without GSO. For example, now that we don't do UFO, you might see significant gains with UDP. If I understand correctly, it can also help for small packet regr. in multiqueue scenario? Well BQL generally should only be active for 1:1 mappings. Would be nice to see the perf. numbers with multi-queue for small packets streams. How about move the BQL patch out of this series? Let's first converge tx interrupt and then introduce it? (e.g with kicking after queuing X bytes?) Sounds good. -- To unsubscribe from this list: send the line unsubscribe netdev in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq->vq); +virtqueue_kick(sq->vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. Okay, let me try to see if this help. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: > Hello: > > We used to orphan packets before transmission for virtio-net. This breaks > socket accounting and can lead serveral functions won't work, e.g: > > - Byte Queue Limit depends on tx completion nofication to work. > - Packet Generator depends on tx completion nofication for the last > transmitted packet to complete. > - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. > > This series tries to solve the issue by enabling tx interrupts. To minize > the performance impacts of this, several optimizations were used: > > - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx > interrupt untile 3/4 pending packets were sent. > - In host side, interrupt coalescing were used to reduce tx interrupts. > > Performance test results[1] (tx-frames 16 tx-usecs 16) shows: > > - For guest receiving. No obvious regression on throughput were > noticed. More cpu utilization were noticed in few cases. > - For guest transmission. Very huge improvement on througput for small > packet transmission were noticed. This is expected since TSQ and other > optimization for small packet transmission work after tx interrupt. But > will use more cpu for large packets. > - For TCP_RR, regression (10% on transaction rate and cpu utilization) were > found. Tx interrupt won't help but cause overhead in this case. Using > more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq->vq); +virtqueue_kick(sq->vq); ? If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. > Changes from RFC V3: > - Don't free tx packets in ndo_start_xmit() > - Add interrupt coalescing support for virtio-net > Changes from RFC v2: > - clean up code, address issues raised by Jason > Changes from RFC v1: > - address comments by Jason Wang, use delayed cb everywhere > - rebased Jason's patch on top of mine and include it (with some tweaks) > > Please reivew. Comments were more than welcomed. > > [1] Performance Test result: > > Environment: > - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back > with 82599ES cards. > - Both host and guest were net-next.git plus the patch > - Coalescing parameters for the card: > Adaptive RX: off TX: off > rx-usecs: 1 > rx-frames: 0 > tx-usecs: 0 > tx-frames: 0 > - Vhost_net was enabled and zerocopy was disabled > - Tests was done by netperf-2.6 > - Guest has 2 vcpus with single queue virtio-net > > Results: > - Numbers of square brackets are whose significance is grater than 95% > > Guest RX: > > size/sessions/+throughput/+cpu/+per_cpu_throughput/ > 64/1/+2.0326/[+6.2807%]/-3.9970%/ > 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ > 64/4/+1.5956%/+2.2451%/-0.6353%/ > 64/8/+1.1732%/+3.5123%/-2.2598%/ > 256/1/+3.7619%/[+5.8117%]/-1.9372%/ > 256/2/-0.0661%/[+3.2511%]/-3.2127%/ > 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ > 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ > 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ > 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ > 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ > 1024/8/+1.1444%/-1.6069%/+2.7961%/ > 4096/1/+0.0401%/-0.5993%/+0.6433%/ > 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ > 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ > 4096/8/-0.3362%/+2.7086%/-2.9645%/ > 16384/1/-0.0285%/+0.7247%/-0.7478%/ > 16384/2/-0.5286%/+0.3287%/-0.8545%/ > 16384/4/-0.3297%/-2.0543%/+1.7608%/ > 16384/8/+1.0932%/+4.0253%/-2.8187%/ > 65535/1/+0.0003%/-0.1502%/+0.1508%/ > 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ > 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ > 65535/8/+1.8359%/+3.1590%/-1.2825%/ > > Guest RX: > size/sessions/+throughput/+cpu/+per_cpu_throughput/ > 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ > 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ > 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ > 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ > 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ > 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ > 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ > 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ > 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/ > 1024/2/[-1.4207%]/[-23.6426%]/[+29.1025%]/ > 1024/4/-0.1003%/[-13.6416%]/[+15.6804%]/ > 1024/8/[+0.2200%]/[+2.0634%]/[-1.8061%]/ > 4096/1/[+18.4835%]/[-46.1508%]/[+120.0283%]/ > 4096/2/+0.1770%/[-26.2780%]/[+35.8848%]/ > 4096/4/-0.1012%/-0.7353%/+0.6388%/ > 4096/8/-0.6091%/+1.4159%/-1.9968%/ > 16384/1/-0.0424%/[+11.9373%]/[-10.7021%]/ > 16384/2/+0.0482%/+2.4685%/-2.3620%/ > 16384/4/+0.0840%/[+5.3587%]/[-5.0064%]/ > 16384/8/+0.0048%/[+5.0176%]/[-4.7733%]/ > 65535/1/-0.0095%/[+10.9408%]/[-9.8705%]/ > 65535/2/+0.1515%/[+8.1709%]/[-7.4137%]/ >
[PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/ 1024/2/[-1.4207%]/[-23.6426%]/[+29.1025%]/ 1024/4/-0.1003%/[-13.6416%]/[+15.6804%]/ 1024/8/[+0.2200%]/[+2.0634%]/[-1.8061%]/ 4096/1/[+18.4835%]/[-46.1508%]/[+120.0283%]/ 4096/2/+0.1770%/[-26.2780%]/[+35.8848%]/ 4096/4/-0.1012%/-0.7353%/+0.6388%/ 4096/8/-0.6091%/+1.4159%/-1.9968%/ 16384/1/-0.0424%/[+11.9373%]/[-10.7021%]/ 16384/2/+0.0482%/+2.4685%/-2.3620%/ 16384/4/+0.0840%/[+5.3587%]/[-5.0064%]/ 16384/8/+0.0048%/[+5.0176%]/[-4.7733%]/ 65535/1/-0.0095%/[+10.9408%]/[-9.8705%]/ 65535/2/+0.1515%/[+8.1709%]/[-7.4137%]/ 65535/4/+0.0203%/[+5.4316%]/[-5.1325%]/ 65535/8/+0.1427%/[+6.2753%]/[-5.7705%]/ size/sessions/+trans.rate/+cpu/+per_cpu_trans.rate/ 64/1/+0.2346%/[+11.5080%]/[-10.1099%]/ 64/25/[-10.7893%]/-0.5791%/[-10.2697%]/ 64/50/[-11.5997%]/-0.3429%/[-11.2956%]/ 256/1/+0.7219%/[+13.2374%]/[-11.0524%]/ 256/25/-6.9567%/+0.8887%/[-7.7763%]/ 256/50/[-4.8814%]/-0.0338%/[-4.8492%]/ 4096/1/-1.6061%/-0.7561%/-0.8565%/ 4096/25/[+2.2120%]/[+1.0839%]/+1.1161%/ 4096/50/[+5.6180%]/[+3.2116%]/[+2.3315%]/ Jason Wang (4): virtio_net: enable tx interrupt virtio-net: optimize free_old_xmit_skbs stats virtio-net: add basic interrupt coalescing support vhost_net: interrupt coalescing support Michael S. Tsirkin (1): virtio_net: bql drivers/net/virtio_net.c| 211
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Mon, Dec 1, 2014 at 6:42 PM, Michael S. Tsirkin m...@redhat.com wrote: On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? Helps a lot. For RX, it saves about 5% - 10% cpu. (reduce 60%-90% tx intrs) For small packet TX, it increases 33% - 245% throughput. (reduce about 60% inters) For TCP_RR, it increase the 3%-10% trans.rate. (reduce 40%-80% tx intrs) I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? Anyway guest need to take some cycles to handle tx interrupts. And transaction rate does increase if we coalesces more tx interurpts. It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? I will try, but during TCP_RR, at most 1 packets were pending, I suspect if BQL can help in this case. If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. Okay, let me try to see if this help. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/
[PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/ 1024/2/[-1.4207%]/[-23.6426%]/[+29.1025%]/ 1024/4/-0.1003%/[-13.6416%]/[+15.6804%]/ 1024/8/[+0.2200%]/[+2.0634%]/[-1.8061%]/ 4096/1/[+18.4835%]/[-46.1508%]/[+120.0283%]/ 4096/2/+0.1770%/[-26.2780%]/[+35.8848%]/ 4096/4/-0.1012%/-0.7353%/+0.6388%/ 4096/8/-0.6091%/+1.4159%/-1.9968%/ 16384/1/-0.0424%/[+11.9373%]/[-10.7021%]/ 16384/2/+0.0482%/+2.4685%/-2.3620%/ 16384/4/+0.0840%/[+5.3587%]/[-5.0064%]/ 16384/8/+0.0048%/[+5.0176%]/[-4.7733%]/ 65535/1/-0.0095%/[+10.9408%]/[-9.8705%]/ 65535/2/+0.1515%/[+8.1709%]/[-7.4137%]/ 65535/4/+0.0203%/[+5.4316%]/[-5.1325%]/ 65535/8/+0.1427%/[+6.2753%]/[-5.7705%]/ size/sessions/+trans.rate/+cpu/+per_cpu_trans.rate/ 64/1/+0.2346%/[+11.5080%]/[-10.1099%]/ 64/25/[-10.7893%]/-0.5791%/[-10.2697%]/ 64/50/[-11.5997%]/-0.3429%/[-11.2956%]/ 256/1/+0.7219%/[+13.2374%]/[-11.0524%]/ 256/25/-6.9567%/+0.8887%/[-7.7763%]/ 256/50/[-4.8814%]/-0.0338%/[-4.8492%]/ 4096/1/-1.6061%/-0.7561%/-0.8565%/ 4096/25/[+2.2120%]/[+1.0839%]/+1.1161%/ 4096/50/[+5.6180%]/[+3.2116%]/[+2.3315%]/ Jason Wang (4): virtio_net: enable tx interrupt virtio-net: optimize free_old_xmit_skbs stats virtio-net: add basic interrupt coalescing support vhost_net: interrupt coalescing support Michael S. Tsirkin (1): virtio_net: bql drivers/net/virtio_net.c| 211
Re: [PATCH RFC v4 net-next 0/5] virtio_net: enabling tx interrupts
On Mon, Dec 01, 2014 at 06:17:03PM +0800, Jason Wang wrote: Hello: We used to orphan packets before transmission for virtio-net. This breaks socket accounting and can lead serveral functions won't work, e.g: - Byte Queue Limit depends on tx completion nofication to work. - Packet Generator depends on tx completion nofication for the last transmitted packet to complete. - TCP Small Queue depends on proper accounting of sk_wmem_alloc to work. This series tries to solve the issue by enabling tx interrupts. To minize the performance impacts of this, several optimizations were used: - In guest side, virtqueue_enable_cb_delayed() was used to delay the tx interrupt untile 3/4 pending packets were sent. - In host side, interrupt coalescing were used to reduce tx interrupts. Performance test results[1] (tx-frames 16 tx-usecs 16) shows: - For guest receiving. No obvious regression on throughput were noticed. More cpu utilization were noticed in few cases. - For guest transmission. Very huge improvement on througput for small packet transmission were noticed. This is expected since TSQ and other optimization for small packet transmission work after tx interrupt. But will use more cpu for large packets. - For TCP_RR, regression (10% on transaction rate and cpu utilization) were found. Tx interrupt won't help but cause overhead in this case. Using more aggressive coalescing parameters may help to reduce the regression. OK, you do have posted coalescing patches - does it help any? I'm not sure the regression is due to interrupts. It would make sense for CPU but why would it hurt transaction rate? It's possible that we are deferring kicks too much due to BQL. As an experiment: do we get any of it back if we do -if (kick || netif_xmit_stopped(txq)) -virtqueue_kick(sq-vq); +virtqueue_kick(sq-vq); ? If yes, we can just kick e.g. periodically, e.g. after queueing each X bytes. Changes from RFC V3: - Don't free tx packets in ndo_start_xmit() - Add interrupt coalescing support for virtio-net Changes from RFC v2: - clean up code, address issues raised by Jason Changes from RFC v1: - address comments by Jason Wang, use delayed cb everywhere - rebased Jason's patch on top of mine and include it (with some tweaks) Please reivew. Comments were more than welcomed. [1] Performance Test result: Environment: - Two Intel(R) Xeon(R) CPU E5620 @ 2.40GHz machines connected back to back with 82599ES cards. - Both host and guest were net-next.git plus the patch - Coalescing parameters for the card: Adaptive RX: off TX: off rx-usecs: 1 rx-frames: 0 tx-usecs: 0 tx-frames: 0 - Vhost_net was enabled and zerocopy was disabled - Tests was done by netperf-2.6 - Guest has 2 vcpus with single queue virtio-net Results: - Numbers of square brackets are whose significance is grater than 95% Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/+2.0326/[+6.2807%]/-3.9970%/ 64/2/-0.2104%/[+3.2012%]/[-3.3058%]/ 64/4/+1.5956%/+2.2451%/-0.6353%/ 64/8/+1.1732%/+3.5123%/-2.2598%/ 256/1/+3.7619%/[+5.8117%]/-1.9372%/ 256/2/-0.0661%/[+3.2511%]/-3.2127%/ 256/4/+1.1435%/[-8.1842%]/[+10.1591%]/ 256/8/[+2.2447%]/[+6.2044%]/[-3.7283%]/ 1024/1/+9.1479%/[+12.0997%]/[-2.6332%]/ 1024/2/[-17.3341%]/[+0.%]/[-17.3341%]/ 1024/4/[-0.6284%]/-1.0376%/+0.4135%/ 1024/8/+1.1444%/-1.6069%/+2.7961%/ 4096/1/+0.0401%/-0.5993%/+0.6433%/ 4096/2/[-0.5894%]/-2.2071%/+1.6542%/ 4096/4/[-0.5560%]/-1.4969%/+0.9553%/ 4096/8/-0.3362%/+2.7086%/-2.9645%/ 16384/1/-0.0285%/+0.7247%/-0.7478%/ 16384/2/-0.5286%/+0.3287%/-0.8545%/ 16384/4/-0.3297%/-2.0543%/+1.7608%/ 16384/8/+1.0932%/+4.0253%/-2.8187%/ 65535/1/+0.0003%/-0.1502%/+0.1508%/ 65535/2/[-0.6065%]/+0.2309%/-0.8355%/ 65535/4/[-0.6861%]/[+3.9451%]/[-4.4554%]/ 65535/8/+1.8359%/+3.1590%/-1.2825%/ Guest RX: size/sessions/+throughput/+cpu/+per_cpu_throughput/ 64/1/[+65.0961%]/[-8.6807%]/[+80.7900%]/ 64/2/[+6.0288%]/[-2.2823%]/[+8.5052%]/ 64/4/[+5.9038%]/[-2.1834%]/[+8.2677%]/ 64/8/[+5.4154%]/[-2.1804%]/[+7.7651%]/ 256/1/[+184.6462%]/[+4.8906%]/[+171.3742%]/ 256/2/[+46.0731%]/[-8.9626%]/[+60.4539%]/ 256/4/[+45.8547%]/[-8.3027%]/[+59.0612%]/ 256/8/[+45.3486%]/[-8.4024%]/[+58.6817%]/ 1024/1/[+432.5372%]/[+3.9566%]/[+412.2689%]/ 1024/2/[-1.4207%]/[-23.6426%]/[+29.1025%]/ 1024/4/-0.1003%/[-13.6416%]/[+15.6804%]/ 1024/8/[+0.2200%]/[+2.0634%]/[-1.8061%]/ 4096/1/[+18.4835%]/[-46.1508%]/[+120.0283%]/ 4096/2/+0.1770%/[-26.2780%]/[+35.8848%]/ 4096/4/-0.1012%/-0.7353%/+0.6388%/ 4096/8/-0.6091%/+1.4159%/-1.9968%/ 16384/1/-0.0424%/[+11.9373%]/[-10.7021%]/ 16384/2/+0.0482%/+2.4685%/-2.3620%/ 16384/4/+0.0840%/[+5.3587%]/[-5.0064%]/ 16384/8/+0.0048%/[+5.0176%]/[-4.7733%]/ 65535/1/-0.0095%/[+10.9408%]/[-9.8705%]/ 65535/2/+0.1515%/[+8.1709%]/[-7.4137%]/ 65535/4/+0.0203%/[+5.4316%]/[-5.1325%]/ 65535/8/+0.1427%/[+6.2753%]/[-5.7705%]/