jmeter-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From William Ottley <williamott...@gmail.com>
Subject Re: simulate test as a LB, with 2 webservers at one time
Date Sun, 04 Jul 2010 18:41:52 GMT
very little difference. there is, but not all that much more noticeable.

this is why i'm very confused with how to test the load balancer.

If I do the EXACT same test I preform on the WS1, onto the LB, then it
should be the same, plus a bit for "another hop"
and this is what I get.. so this is fine.

now this is with 300 users. so, if I have 2 x WS, WS1 and WS2, identical
machines, returning identical results,
then in THEORY, I should be able to send 600 users to the LB, that will
evenly send them out to  WS1 and WS2, which I have done in the config.
I even confirm that there are the same amount of connections for each WS:

watch ipvsadm --list --stats

and well the results, SHOULD be the same latency + 1 hop worth of latency as
a single WS, right?
I don't.. I get 2x the latency, OR, the same latency as I would if I sent
600 users to a single WS...

SO, since this is the case, I want to rule out the LB, and in order to do
that, I have to do a test, that would simulate the LB with my JMETER.

So the question is how do I do that?

I thought, logically I would do 600 users, but that produces the same
results as sending 600 users to only 1 of the WS...

so I'm a bit confused, and so I thought I'd ask someone if they have done a
test like this, and what's the best way to approach..

thanks for your time!

Will



On Sun, Jul 4, 2010 at 2:23 PM, Sudip Kumar Bhattacharya <
skb.subscriptions@gmail.com> wrote:

> Check your web server logs to verify that the loads are getting equally
> distributed between the 2 servers.
> Also the diff between the 2 modes is that first u had direct access to the
> 2 servers, but now u r going thu an extra hop on the LB.
> This extra hop will also add to the latency.
> Try this. Target ws1 directly with 600 hits, and target ws1 again with 600
> hits but this time thru the lb. See how much latency the lb adds.
> Sent on my BlackBerry® from Vodafone
>
> -----Original Message-----
> From: William Ottley <williamottley@gmail.com>
> Date: Sun, 4 Jul 2010 14:16:53
> To: JMeter Users List<jmeter-user@jakarta.apache.org>
> Reply-To: "JMeter Users List" <jmeter-user@jakarta.apache.org>
> Subject: Re: simulate test as a LB, with 2 webservers at one time
>
> Your right.
> When I test the LB: 600 users, with 3 httpdclient urls.
> I test this to see if I get "close" to the same latency, as 2 web
> servers getting 300 users each with 3 httpdclient urls.
>
> I don't. I get 2x the latency, or the same latency as when I send 600
> users with 3 urls to 1 webserver.
>
> So I'm trying to find where the problem is.
>
> Sent from my iPhone
>
> On 2010-07-04, at 1:57 PM, "Sudip Kumar Bhattacharya" <
> skb.subscriptions@gmail.com
>  > wrote:
>
> > If you are using a load balancer it wud normally front end on all
> > requests.
> > You wud not have to generate different urls on jmeter to reach
> > different servers as your mail suggests.
> > Having said that. Load balancers also have stickiness rules which
> > send consecutive requests from the same session or same ip to the
> > same web server for maintaining session data continuity.
> > You may not get an equally distributed load balancing between the
> > two servers.
> > Chen out your particular LB algorithm setting to get a better idea.
> > Sent on my BlackBerry® from Vodafone
> >
> > -----Original Message-----
> > From: William Ottley <williamottley@gmail.com>
> > Date: Sun, 4 Jul 2010 13:47:59
> > To: <jmeter-user@jakarta.apache.org>
> > Reply-To: "JMeter Users List" <jmeter-user@jakarta.apache.org>
> > Subject: simulate test as a LB, with 2 webservers at one time
> >
> > Hi all,
> >
> > I'm trying to solve an issue with regards to latency and Load
> > Balancing.
> >
> > I have done several latency tests using jmeter.
> > I have tested each of the webservers with this:
> >
> > 300 threads, and 3x HHTP Request HTTPClient, each one has a
> > different url
> > location.
> >
> > If I wanted to get the results of a Load Balancer, serving the 2
> > webservers,
> > would I:
> >
> > a) do 600 threads, and 3x HHTP Request HTTPClient, each one has a
> > different
> > url location (for webserver 1) and 3x HTTP Request HTTPClient (for
> > webserver
> > 2)
> > or
> > b) do 300 threads and 6x HTTP Request HTTPClient, done so that the
> > first one
> > is with WS1, first url, 2nd one is WS2, first url, etc....
> >
> > in my head, logically, a LB, should be able to sever 600 clients,
> > with 3
> > HTTPClient requests: while having almost the same latency.. but i'm
> > not
> > getting that.
> >
> >
> > so I decided to use jmeter to simulate conecting to both webservers
> > at the
> > same time...
> >
> > I hope this makes sense....
> >
> >
> > thanks!
> >
> > William
> >
>
> ---------------------------------------------------------------------
> 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