Netty Warm Up Stage

[Arvid Svensson]

Double check the code calculating your numbers. How and when do you start
counting messages and time? The JVM warm up a lot faster than your numbers.

Rgds,
Arvid

On Thu, Oct 27, 2011 at 10:28 PM, Norman Maurer <
norman.maurer at googlemail.com> wrote:

> Also it would help if you could show us your code ;)
>
> Bye,
> Norman
>
>
> 2011/10/27 Norman Maurer <norman.maurer at googlemail.com>:
> > Also netty use a pool for the buffers. I'm not remember at the moment
> > if its "populated" in a lazy fashion or not.
> >
> > Bye,
> > Norman
> >
> >
> > 2011/10/27 John D. Mitchell <jdmitchell at gmail.com>:
> >> Just to be clear, that doesn't fully explain what the OP is seeing.
> Typically, I see things warmed up on the hot paths in the 10K-100K
> iterations through that path. The OP is sending 10MM messages so just jvm
> warm up issues wouldn't explain the performance curve that they are seeing.
> >>
> >> John
> >>
> >> On Oct 27, 2011, at 13:20 , Norman Maurer wrote:
> >>
> >>> Exactly... The JVM needs some warmup for best performance.
> >>>
> >>> Bye,
> >>> Norman
> >>>
> >>>
> >>> 2011/10/27 John D. Mitchell <jdmitchell at gmail.com>:
> >>>> There's the JVM dynamic compilation / optimization stages and things
> like the TCP slow-start.
> >>>>
> >>>> The only way to get over the JVM start up is to run your code for
> awhile. Given the number of messages you're sending, I think you're doing
> plenty to warm that part up.
> >>>>
> >>>> TCP issues are done on a per-connection basis so there's not a lot you
> can do there unless you can control the settings of the machines on both
> ends and the routers in between. But at least you can tweak the various
> settings on the server that your netty code is running on. Alas, there are
> lots of tradeoffs in this so there's not a simplistic one-size-fits-all
> answer.
> >>>>
> >>>> Hope this helps,
> >>>> John
> >>>>
> >>>> On Oct 27, 2011, at 11:34 , zx spectrum wrote:
> >>>>
> >>>>> When I send / receive 10,000,000 messages to try to measure the rate,
> I see that Netty spends some time to warm up:
> >>>>>    consuming rate: 64800 messages/s
> >>>>>    consuming rate: 106016 messages/s
> >>>>>    consuming rate: 143905 messages/s
> >>>>>    consuming rate: 177714 messages/s
> >>>>>    consuming rate: 203981 messages/s
> >>>>>    consuming rate: 228632 messages/s
> >>>>>    consuming rate: 252397 messages/s
> >>>>>    consuming rate: 272507 messages/s
> >>>>>    consuming rate: 291819 messages/s
> >>>>>    consuming rate: 309933 messages/s
> >>>>>
> >>>>> This is the average rate of the messages ( that pass through
> FrameDecoder ) received per each million.
> >>>>>
> >>>>> Can anybody explain why it happens? And how to eliminate a "warm up"
> stage?
> >>>>>
> >>>>> Thank you!!
> >>>>>
> >>>>> _______________________________________________
> >>>>> netty-users mailing list
> >>>>> netty-users at lists.jboss.org
> >>>>> https://lists.jboss.org/mailman/listinfo/netty-users
> >>>>
> >>>>
> >>>> _______________________________________________
> >>>> netty-users mailing list
> >>>> netty-users at lists.jboss.org
> >>>> https://lists.jboss.org/mailman/listinfo/netty-users
> >>>>
> >>>
> >>> _______________________________________________
> >>> netty-users mailing list
> >>> netty-users at lists.jboss.org
> >>> https://lists.jboss.org/mailman/listinfo/netty-users
> >>
> >>
> >> _______________________________________________
> >> netty-users mailing list
> >> netty-users at lists.jboss.org
> >> https://lists.jboss.org/mailman/listinfo/netty-users
> >>
> >
>
> _______________________________________________
> netty-users mailing list
> netty-users at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/netty-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.jboss.org/pipermail/netty-users/attachments/20111027/22378d2b/attachment.html