Re: [ovs-dev] ovs-vswitchd 2.4.1 scale >10K add/delete flows 100% cpu
On Wed, Jul 11, 2018 at 10:33:51AM -0700, Ravi Kerur wrote: > On Tue, Jul 10, 2018 at 2:07 PM Ben Pfaff wrote: > > > On Thu, Jul 05, 2018 at 01:44:33PM -0700, Ravi Kerur wrote: > > > During scale flow add/delete (>10K), I am seeing ovs-vswitchd cpu usage > > > spike to 100% and stay there without any sign of returning to normal cpu > > > usage. It's normal OVS 2.4.1 and no DPDK involved. I am trying to get > > > 'perf' working which might help in isolating the problem. In the > > meantime I > > > would like to understand following things > > > > > > (1) Recommended system configuration i.e. core allocation, memory, > > > hugepages, ... > > > (2) Published scale numbers for 2.4.1 > > > (3) Known performance issues with 2.4.1 > > > (4) Debugs to collect > > > > 2.4.x is really old. It was released in 2015. I doubt anyone is going > > to be able to help you with it. > > > > Thanks for the information. We are using OF1.3, (10K) Add/Delete come into > OVS as a individual flowmod. > > Currently I have three versions in our lab > > (1) OVS 2.4.1 -- issue is seen, I am still trying to get 'perf' running on > that system. Once I have it running I will post more details. > (2) OVS 2.5.x -- Using this version with DPDK for other things > (3) OVS 2.9.90 -- Currently testing for vhost-virtio functionality > > Do you think OVS 2.5.x might be better suited for large number of > individual flowmods or you recommend 2.9.90? Usually I recommend the most recent version of OVS that you're comfortable with. ___ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
Re: [ovs-dev] ovs-vswitchd 2.4.1 scale >10K add/delete flows 100% cpu
On Tue, Jul 10, 2018 at 2:07 PM Ben Pfaff wrote: > On Thu, Jul 05, 2018 at 01:44:33PM -0700, Ravi Kerur wrote: > > During scale flow add/delete (>10K), I am seeing ovs-vswitchd cpu usage > > spike to 100% and stay there without any sign of returning to normal cpu > > usage. It's normal OVS 2.4.1 and no DPDK involved. I am trying to get > > 'perf' working which might help in isolating the problem. In the > meantime I > > would like to understand following things > > > > (1) Recommended system configuration i.e. core allocation, memory, > > hugepages, ... > > (2) Published scale numbers for 2.4.1 > > (3) Known performance issues with 2.4.1 > > (4) Debugs to collect > > 2.4.x is really old. It was released in 2015. I doubt anyone is going > to be able to help you with it. > Thanks for the information. We are using OF1.3, (10K) Add/Delete come into OVS as a individual flowmod. Currently I have three versions in our lab (1) OVS 2.4.1 -- issue is seen, I am still trying to get 'perf' running on that system. Once I have it running I will post more details. (2) OVS 2.5.x -- Using this version with DPDK for other things (3) OVS 2.9.90 -- Currently testing for vhost-virtio functionality Do you think OVS 2.5.x might be better suited for large number of individual flowmods or you recommend 2.9.90? ___ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
Re: [ovs-dev] ovs-vswitchd 2.4.1 scale >10K add/delete flows 100% cpu
On Thu, Jul 05, 2018 at 01:44:33PM -0700, Ravi Kerur wrote: > During scale flow add/delete (>10K), I am seeing ovs-vswitchd cpu usage > spike to 100% and stay there without any sign of returning to normal cpu > usage. It's normal OVS 2.4.1 and no DPDK involved. I am trying to get > 'perf' working which might help in isolating the problem. In the meantime I > would like to understand following things > > (1) Recommended system configuration i.e. core allocation, memory, > hugepages, ... > (2) Published scale numbers for 2.4.1 > (3) Known performance issues with 2.4.1 > (4) Debugs to collect 2.4.x is really old. It was released in 2015. I doubt anyone is going to be able to help you with it. ___ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
[ovs-dev] ovs-vswitchd 2.4.1 scale >10K add/delete flows 100% cpu
Hi, During scale flow add/delete (>10K), I am seeing ovs-vswitchd cpu usage spike to 100% and stay there without any sign of returning to normal cpu usage. It's normal OVS 2.4.1 and no DPDK involved. I am trying to get 'perf' working which might help in isolating the problem. In the meantime I would like to understand following things (1) Recommended system configuration i.e. core allocation, memory, hugepages, ... (2) Published scale numbers for 2.4.1 (3) Known performance issues with 2.4.1 (4) Debugs to collect ovs-ofctl --version ovs-ofctl (Open vSwitch) 2.4.1 Compiled May 25 2016 13:31:48 OpenFlow versions 0x1:0x4 PID USER PR NIVIRTRESSHR S %CPU %MEM TIME+ COMMAND 56446 root 20 0 45.894g 0.031t 12084 S 814.7 12.6 788147:01 qemu-system-x86 28509 root 10 -10 2215740 310668 7304 R 100.2 0.1 10502:22 ovs-vswitchd 55579 root 20 0 45.688g 0.029t 12508 S 15.8 11.8 511176:21 qemu-system-x86 Enabled some debugs ovs-appctl vlog/set poll_loop:DBG and logs show 2018-07-04T14:27:20.307Z|00474|connmgr|INFO|vn-vn9014<->unix: 17000 flow_mods in the last 56 s (16612 adds, 388 deletes) 2018-07-04T14:27:24.213Z|00475|connmgr|INFO|vn-vn9046<->unix: 17188 flow_mods in the last 58 s (16610 adds, 578 deletes) 2018-07-04T14:28:20.306Z|00476|connmgr|INFO|vn-vn9014<->unix: 2786 flow_mods in the last 59 s (2786 adds) 2018-07-04T14:28:24.214Z|00477|connmgr|INFO|vn-vn9046<->unix: 2790 flow_mods in the last 59 s (2790 adds) 2018-07-04T14:28:30.002Z|00478|poll_loop|INFO|Dropped 48 log messages in last 151 seconds (most recently, 149 seconds ago) due to excessive rate 2018-07-04T14:28:30.002Z|00479|poll_loop|INFO|wakeup due to [POLLIN] on fd 446 (FIFO pipe:[2524722865]) at ofproto/ofproto-dpif.c:1574 (*58% CPU usage)* 2018-07-04T14:28:30.002Z|00480|poll_loop|INFO|wakeup due to [POLLIN] on fd 453 (/var/run/openvswitch/vn-vn9014.mgmt<->) at *lib/stream-fd.c:155 (58% CPU usage)* 2018-07-04T14:28:30.002Z|00481|poll_loop|INFO|wakeup due to [POLLIN] on fd 453 (/var/run/openvswitch/vn-vn9014.mgmt<->) at lib/stream-fd.c:155 (58% CPU usage) 2018-07-04T14:28:30.002Z|00482|poll_loop|INFO|wakeup due to 0-ms timeout at ofproto/ofproto-dpif.c:1571 (*58% CPU usage*) 2018-07-04T14:28:30.002Z|00483|poll_loop|INFO|wakeup due to [POLLIN] on fd 446 (FIFO pipe:[2524722865]) at lib/ovs-rcu.c:206 (*58% CPU usage*) 2018-07-04T14:28:30.003Z|00484|poll_loop|INFO|wakeup due to 0-ms timeout at lib/ovs-rcu.c:206 (58% CPU usage) 2018-07-04T14:28:30.003Z|00485|poll_loop|INFO|wakeup due to [POLLIN] on fd 446 (FIFO pipe:[2524722865]) at lib/ovs-rcu.c:206 (*58% CPU usage*) ... Eventually CPU usage reaches 100% and stays there. Thanks. ___ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev