Re: [dpdk-dev] [PATCH 24/33] app/testeventdev: add perf queue test

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

-----Original Message-----
> Date: Fri, 23 Jun 2017 12:47:50 +0000
> From: "Van Haaren, Harry" <harry.van.haaren@intel.com>
> To: Jerin Jacob <jerin.jacob@caviumnetworks.com>, "dev@dpdk.org"
>  <dev@dpdk.org>
> CC: "Richardson, Bruce" <bruce.richardson@intel.com>,
>  "hemant.agrawal@nxp.com" <hemant.agrawal@nxp.com>, "Eads, Gage"
>  <gage.eads@intel.com>, "nipun.gupta@nxp.com" <nipun.gupta@nxp.com>,
>  "Vangati, Narender" <narender.vangati@intel.com>, "Rao, Nikhil"
>  <nikhil.rao@intel.com>, "gprathyusha@caviumnetworks.com"
>  <gprathyusha@caviumnetworks.com>
> Subject: RE: [dpdk-dev] [PATCH 24/33] app/testeventdev: add perf queue test
> 
> > From: Jerin Jacob [mailto:jerin.jacob@caviumnetworks.com]
> > Sent: Sunday, May 28, 2017 8:59 PM
> > To: dev@dpdk.org
> > Cc: Richardson, Bruce <bruce.richardson@intel.com>; Van Haaren, Harry
> > <harry.van.haaren@intel.com>; hemant.agrawal@nxp.com; Eads, Gage <gage.eads@intel.com>;
> > nipun.gupta@nxp.com; Vangati, Narender <narender.vangati@intel.com>; Rao, Nikhil
> > <nikhil.rao@intel.com>; gprathyusha@caviumnetworks.com; Jerin Jacob
> > <jerin.jacob@caviumnetworks.com>
> > Subject: [dpdk-dev] [PATCH 24/33] app/testeventdev: add perf queue test
> > 
> > This is a performance test case that aims at testing the following:
> > 1. Measure the number of events can be processed in a second.
> > 2. Measure the latency to forward an event.
> > 
> > The perf queue test configures the eventdev with Q queues and P ports,
> > where Q is nb_producers * nb_stages and P is nb_workers + nb_producers.
> > 
> > The user can choose the number of workers, the number of producers and
> > number of stages through the --wlcores , --plcores and the --stlist
> > application command line arguments respectively.
> > 
> > The producer(s) injects the events to eventdev based the
> > first stage sched type list requested by the user through --stlist
> > the command line argument.
> > 
> > Based on the number of stages to process(selected through --stlist),
> > the application forwards the event to next upstream queue and
> > terminates when it reaches the last stage in the pipeline.
> > On event termination, application increments the number events
> > processed and print periodically in one second to get the
> > number of events processed in one second.
> 
> We should forward events for longer than one second. From a SW PMD perspective, it takes some time for caches to warm up, so running a test for 1 second isn't very representative. Suggestion of running default 5 seconds, as a balance between consistent performance and keeping the test runs short?

Done. Change the default to 64M packets.

-opt->nb_pkts = (1ULL << 22);
+opt->nb_pkts = (1ULL << 26); /* do ~64M packets */

> 
> I'm aware of the --nb_pkts=X arg, this is more about solid defaults :)
> 
> 
> > When --fwd_latency command line option selected, the application
> > inserts the timestamp in the event on the first stage and then
> > on termination, it updates the number of cycles to forward
> > a packet. The application uses this value to compute the average
> > latency to a forward packet.
> > 
> > Example command to run perf queue test:
> > sudo build/app/dpdk-test-eventdev --vdev=event_sw0 -- --test=perf_queue\
> > --slcore=1 --plcores=2 --wlcore=3 --stlist=p --nb_pkts=1000000000
> > 
> > Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> 
> With the above open;
> 
> Acked-by: Harry van Haaren <harry.van.haaren@intel.com>