jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Deepak Shetty <shet...@gmail.com>
Subject Re: Test plan for (970 page requests every 5 min)
Date Tue, 21 Sep 2010 00:16:39 GMT
>Also, how does a tester calculate the page response time using load
balancer logs?
Its the response time as the server sees it. Your IIS would have logs which
state when the request started and ended as should your Load Balancer. This
ignores network time between client and server (which is variable anyway)
and render time (which also varies). This is a good enough value in most
cases since you can estimate the other two factors.

>And what are those reverse proxy logs he was talking about?
Java based applications have a WebServer which proxies requests to the
actual application servers. Its called a reverse proxy because the direction
is from the internet into your network.

>Yes, but if the requirement states that pages need to onload in less then
10 seconds with 30 concurrent users,
But you dont need accurate times right. You only need to know whether its
less or more.
You can figure out how much network time would be taken (given bandwidth and
size of files since the size doesnt change). You can figure out how much
time representative browsers take to render your page. You can see from a
browser what time is needed for static files download (cached and non
cached). And you might come up with say 2 seconds is the time for all the
other resources (like images CSS/JS/images) without having run a single load
test. You might add a safety margin(say 1 second) and say that your
requirement is actually the main page should be less than 7 seconds with 30
concurrent users . If when you run your tests your actual figures come say 2
seconds for 30 concurrent users you can safely assume you will meet your
time(you might run another test with embedded resources on just to verify ).
If however your values are close to 7 seconds then you may see if your
static file estimate is accurate or not. If the values are higher than 7
seconds then you arent meeting your times in any case (and somedeveloper
will have to tune the app more) so why worry about the static files since
you have to retest anyway?

>Do you mean spikes like these
Spikes can happen (for e.g. if your ramp up time is too less ) usually the
duration of the spike must be more for you to be sure that the server is
being loaded beyond what is good.

On Mon, Sep 20, 2010 at 3:33 PM, Prostak <strodion@gmail.com> wrote:

>
> >a. this is my opinion, use it if it works for you or makes sense to you.
> >Felix for e.g. has a different perspective and you might find something
> >different that makes life easier for you.
>
> I asked Felix, but he disappeared without answering. Do you have any idea
> how Felix measures #3 "The time your browser takes until you can see
> anything, at least some text".
>
> Also, how does a tester calculate the page response time using load
> balancer
> logs?
>
> And what are those reverse proxy logs he was talking about?
>
> >b. very accurate page load times are rarely needed .
>
> Yes, but if the requirement states that pages need to onload in less then
> 10
> seconds with 30 concurrent users, what is the way to measure it, if you
> can't get accurate page load time?
>
> >I would spend more time on the dynamically generated stuff since that's
> >where problems crop up.
>
> Our site doesn't have any dynamically generated stuff. What should I test
> for then, if page load is out of the option?
>
> >If memory serves me correctly you had a load , but the times for the same
> >resource varied wildly. if this is not the case ignore what I said. If the
> >problem is server CPU then this should be easy to verify, run perfmon(or
> >equivalent) while your test is running and check if the CPU is pegged at
> >100% on the server.
>
> Do you mean spikes like these:
>
> http://tinypic.com?ref=2r3ktxk  http://i56.tinypic.com/2r3ktxk.png
>
> Thanks,
> Prostak
> --
> View this message in context:
> http://jmeter.512774.n5.nabble.com/Test-plan-for-970-page-requests-every-5-min-tp2826174p2847310.html
> Sent from the JMeter - User mailing list archive at Nabble.com.
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: jmeter-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: jmeter-user-help@jakarta.apache.org
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message