On Mon, Apr 03, 2017 at 11:16:47AM -0400, Aaron Conole wrote:
> "Bodireddy, Bhanuprakash" <bhanuprakash.bodire...@intel.com> writes:
>
> > Thanks Aaron for reviewing this patch series. My comments inline.
> >
> >>
> >>Hi Bhanu,
> >>
> >>Bhanuprakash Bodireddy <bhanuprakash.bodire...@intel.com> writes:
> >>
> >>> This patch is aimed at achieving Fastpath Service Assurance in
> >>> OVS-DPDK deployments. This commit adds support for monitoring the
> >>> packet processing cores(pmd thread cores) by dispatching heartbeats at
> >>> regular intervals. Incase of heartbeat miss the failure shall be
> >>> detected & reported to higher level fault management
> >>systems/frameworks.
> >>>
> >>> The implementation uses POSIX shared memory object for storing the
> >>> events that will be read by monitoring framework. keep-alive feature
> >>> can be enabled through below OVSDB settings.
> >>>
> >>> keepalive=true
> >>> - Keepalive feature is disabled by default
> >>>
> >>> keepalive-interval="50"
> >>> - Timer interval in milliseconds for monitoring the packet
> >>> processing cores.
> >>>
> >>> keepalive-shm-name="/dpdk_keepalive_shm_name"
> >>> - Shared memory block name where the events shall be updated.
> >>>
> >>> When KA is enabled, 'ovs-keepalive' thread shall be spawned that wakes
> >>> up at regular intervals to update the timestamp and status of pmd
> >>> cores in shared memory region.
> >>>
> >>> An external monitoring framework like collectd(with dpdk plugin
> >>> support) can read the status updates from shared memory. On a missing
> >>> heartbeat, the collectd shall relay the status to ceilometer service
> >>> running in the controller. Below is the high level overview of deployment
> >>model.
> >>
> >>Given this runs in-line in the fastpath, can you tell me what kind of
> >>impact it is
> >>to throughput (and possibly latency) that this series implies?
> >
> > This feature has very minimal impact on the throughput.
> > As you know, there is a significant drop on the current master due to
> > commit: daf4d3c18da4("odp: Support conntrack orig tuple key.").
> > I folded in the RFC patch by Daniele that fixes the memset and measured the
> > throughput with and without KA feature and see no significant difference in
> > throughput.
> >
> > Agree that Latency is indeed important. I would collect the latency
> > stats and will share
> > the results in this thread.
>
> Awesome. Even better would be to put those informations in the cover
> letter of v2. Just a short summary of the tests and what the results
> were for latency / throughput would be good.
I think it'd be even better to include measurements in one of the commit
messages, because those are available in the repository after the
patches are applied. It's harder to find cover letters because they're
only on the mailing list.
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
