Speed difference under load

Chris Lamprecht clamprecht at gmail.com
Sat Oct 27 05:28:23 UTC 2007


Maybe it's just opening the socket to memcached for the first time after
being idle?  One of the Java clients (maybe both) has a socket connection
pool, and I could see this being the case.

On 10/26/07, K J <sanbat at gmail.com> wrote:
>
>   Depending on the language and runtime, you could almost count on it
> > while libraries load and initialize and jits optimize.
> >
> >
> > Testers usually include some ramp-up time where things sort of warm-up
> > before you actually start capturing metrics.
> >
>
> So it's normal to have Memcache's response time be 1 second, if the
> system is relatively idle?  This happens when I'm basically the only user on
> the system, logging in, checking pages, etc.
>
>
>
>
>
> On 10/27/07, Dustin Sallings <dustin at spy.net> wrote:
> >
> >
> >
> >   Depending on the language and runtime, you could almost count on it
> > while libraries load and initialize and jits optimize.
> >
> >
> >   Testers usually include some ramp-up time where things sort of warm-up
> > before you actually start capturing metrics.
> >
> > --  Dustin Sallings (mobile)
> >
> > On Oct 26, 2007, at 20:24, "K J" <sanbat at gmail.com> wrote:
> >
> >
> >
> >  Ok I tested this, and found that the very first memcache get request
> > always takes up 1 second.  All subsequent gets in the same page are fast.
> > Does anyone know what could cause the first memcache get request to slow
> > down like this?
> >
> >
> >
> >
> >
> >
> >
> >
> > On 10/25/07, John Kramlich < <john.kramlich at gigoit.org>john.kramlich at gigoit.org>
> > wrote:
> > >
> > > J A wrote:
> > > > In my application I've noticed something weird.  When testing it
> > > with
> > > > one user, the memcached page loads in like 1 second, but when I use
> > > a
> > > > load testing software and put 20 simultaneous users on it, the page
> > > > load goes to 0.01 second, even though the server load has gone up.
> > > >
> > > > I'm puzzled by this.  Anyone have similar experiences?
> > > >
> > >
> > > What load testing software are you using and what happens if you
> > > configure that load testing software to emulate only a single
> > > user?  If
> > > you are running something like siege locally on your server then
> > > network
> > > latency would be much lower than if you were running the same program
> > > on
> > > a remote machine.  That may be what makes the page load more slowly.
> > >
> > > You may also want to profile you code and see how long returning
> > > results
> > > from memcache takes.  If using PHP you can get a benchmarking class
> > > from
> > > PEAR and set start and stop markers before and after you memcache
> > > related code.  Then you can output the number of milliseconds it takes
> > > to execute that code.  You can do this for other parts of your code as
> > > well.  It certainly helps when figuring out where to spend time
> > > optimizing.
> > >
> > > - John Kramlich
> > > ------------------------------
> > >  <http://www.gigoit.org/>http://www.gigoit.org  - Give and get free
> > > items within your community
> > >
> >
> >
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.danga.com/pipermail/memcached/attachments/20071027/2e93f32d/attachment.htm


More information about the memcached mailing list