Re: [squid-users] Evaluating SQUID as a Reverse Proxy for a large web site

From: kapil khanna <kapil@dont-contact.us>
Date: Sun, 7 Mar 2004 18:20:35 -0600

Koji,
My JMeter plan was as follows.
Access small number of static/dynamic files less than a thousand.
In first case Apache Served up both the static and dynamic content
In second case Squid served up Static content files and Apache served up
Dynamic content.
Apache scaled much better than Squid.
However going by your explanation, since the no of files were less than a
thousand, Apache performed much better due to OS level caching. How many
static files does my test need to be before i start seeing SQUID perform
much better? Is there any way to have SQUID run more than one process/thread
to handle concurrency much better?
Also, are there any benchmarks out there showing what kind of performance
improvement i can hope to expect, 15%, 20 % or more?
Thanks,
--Kapil

----- Original Message -----
From: "Koji Hino" <hino@sv.nec-labs.com>
To: <kapil@j2eerules.com>
Cc: <hno@squid-cache.org>; <squid-users@squid-cache.org>
Sent: Sunday, March 07, 2004 4:23 PM
Subject: Re: [squid-users] Evaluating SQUID as a Reverse Proxy for a large
web site

> Kapil,
>
> Ok, then, how you configured JMeter's testplan?
>
> (Sorry, my previous question was not so well organized..)
>
> (1) Access dynamic contents and static contents, trying to simulate
> real environment. Your Apache server is busy in both case (Apache
> benchmark and Squid benchmark) due to dynamic contents
> generation. Then, you compared results of static content retrieval
> throughput part (you excluded dynamic content retrieval
> throughput part from total throughput results).
>
> (2) Access huge number of static contents.
>
> (3) Access small number (say less than few hundreds) of static
> contents.
>
> If your testplan is (3), Apache's total processing amount (including
> OS's) may be less than Squid's. On Apache server, its OS can cache all
> static content files on memory, and the OS knows those files were not
> modified (because no one write to those files), so the OS and Apache
> are free from many complex checkings and cache managements.
>
> Koji
>
> From: "kapil khanna" <kapil@j2eerules.com>
> Date: Sun, 7 Mar 2004 15:36:04 -0600
>
> :> Koji,
> :> i benchmarked using JMeter a open source tool for generating Http
Requests.
> :> Since this was a controlled benchmark, i guess it would be considered
> :> synthetic.
> :> Thanks,
> :> --Kapil
> :> ----- Original Message -----
> :> From: "Koji Hino" <hino@sv.nec-labs.com>
> :> To: <kapil@j2eerules.com>
> :> Cc: <hno@squid-cache.org>; <squid-users@squid-cache.org>
> :> Sent: Sunday, March 07, 2004 3:22 PM
> :> Subject: Re: [squid-users] Evaluating SQUID as a Reverse Proxy for a
large
> :> web site
> :>
> :>
> :> > Hi,
> :> >
> :> > How you "benchmark" Apache and Squid?
> :> >
> :> > I mean:
> :> >
> :> > * Real environment benchmark, i.e. compare between followings with
real
> :> > customers
> :> > - Get static contents from your busy (due to dynamic contents
> :> > generation) Apache server
> :> > - Get static contents from Squid, and get dynamic contents from
your
> :> > Apache server through Squid
> :> >
> :> > OR
> :> >
> :> > * Synthetic benchmark such as:
> :> > - Get limited number (say several hundreds) of static contents,
from
> :> > your Apache server, or from your Squid server, with benchmarking
> :> > HTTP client(s)
> :> >
> :> > I think those two benchmarks are completely different things.
> :> >
> :> > Best regards,
> :> >
> :> > ====================================================================
> :> > Koji HINO(HINO is my family name)
> :> > NEC Laboratories America, Inc.
> :> >
>
Received on Sun Mar 07 2004 - 17:20:43 MST

This archive was generated by hypermail pre-2.1.9 : Thu Apr 01 2004 - 12:00:01 MST