You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@tomcat.apache.org by Sean Dockery <us...@sbdconsultants.com> on 2003/12/15 09:48:49 UTC
Tomcat performance on Windows versus Linux
I am planning to profile a web application on Windows XP (my development
platform). I am curious as to whether or not different components in Tomcat
and the JVM will behave differently (in a relative comparison) on Linux
(production platform) than Windows.
For example, I have had a person tell me that threads under Linux are more
performant than threads under Windows--leading to the corollary that web
applications under Linux are more performant than web applications under
Windows on the same hardware. My guess is that this claim is based upon the
supposition that thread/context switches under Linux are faster than under
Windows. I find the claim rather dubious because I've never seen data to
support the claim, but doubt is not certainty.
Is there any evidence that this claim and other component performance
differences between the Windows and Linux platform exist and are significant
enough to throw my performance measurements out the "window". :-)
My concern is that I'll profile the application under Windows and tune it,
but then find that my gains aren't as significant or maybe even worthless
under Linux.
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Tomcat performance on Windows versus Linux
Posted by Sean Dockery <us...@sbdconsultants.com>.
"David Rees" <dr...@greenhydrant.com> wrote in message
news:1865.208.48.139.163.1071515435.squirrel@www.greenhydrant.com...
> Not enough difference to make it a deciding factor between the two
> platforms. IMO, Tim's criteria are spot on when deciding what platform to
> deploy on. Personally, I prefer Unix as I find it easier to setup and
> administer. Of course, the majority of my experience with Tomcat is on
> Unix, and not on Windows.
That's fine. I never disagreed with Tim's reasons to choose one platform
over another. But, as I have already responded, I am not choosing a
platform on which to deploy my application; so, arguments as to which
platform is better are moot.
> If you haven't looked already, have a look at the Volano benchmarks
> (google for it) for some numbers on the scalability and performance of
> different JVM, but note that those numbers won't necessarily reflect the
> performance of YOUR application running on Tomcat.
Thanks for this tidbit. It is still very early in the process, but
scalability of JVMs is definitely of interest to me.
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Tomcat performance on Windows versus Linux
Posted by David Rees <dr...@greenhydrant.com>.
On Mon, December 15, 2003 at 9:42 am, Sean Dockery wrote:
> "Tim Funk" <fu...@joedog.org> wrote in message:
>> [I hate saying this since its rather very much like flambait
>> but...]
>>
>> If its worth anything, I haven't had enough load on any of our apps
>> to know whether Linux or Windows is better. Instead, look at: ***
>> - Maintenance - If your a windows shop - stay windows *** -
>> Debugging - I think troubleshooting is easier on *nix systems
>> (YMMV) - Comfort - If your comfortable with unix concepts - linux
>> might be easier than windows
>
> Thanks, Tim, for the even handed response.
>
> I'm not looking for a business case to choose one or the other,
> however; it is certain that our customers will be deploying our
> application on both Linux and Windows (and even Solaris). I'm just
> looking to find out whether or not OS service (TCP/IP stacks,
> threads, file I/O, etc...) implementation differences between Linux
> and Windows have a significant impact on performance and thus should
> be weighed accordingly.
Not enough difference to make it a deciding factor between the two
platforms. IMO, Tim's criteria are spot on when deciding what platform to
deploy on. Personally, I prefer Unix as I find it easier to setup and
administer. Of course, the majority of my experience with Tomcat is on
Unix, and not on Windows.
If you haven't looked already, have a look at the Volano benchmarks
(google for it) for some numbers on the scalability and performance of
different JVM, but note that those numbers won't necessarily reflect the
performance of YOUR application running on Tomcat.
> My conclusions from my readings so far: Slow java code (i.e.:
> algorithms) will be slow on any platform; change the implementation
> to make it faster. Configurable behaviour dependent upon OS services
> (TCP/IP stacks, threads, file I/O, etc...) should be tuned for the
> platform on which the application will live.
I think you've got the idea.
-Dave
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Tomcat performance on Windows versus Linux
Posted by Oscar Carrillo <to...@daydream.stanford.edu>.
I am about to setup Tomcat under a new Linux 2.6 kernel with 2 Athlon MP
processors. Since scheduling, threading, and SMP have been much improved
in the new kernel I wonder if it will add to performance.
I don't have anything to test the new setup with, but if anyone has good
ideas (and by good, I mean "easy"), as I haven't done any profiling, etc.
Oscar
http://daydream.stanford.edu/tomcat/install_web_services.html
On Mon, 15 Dec 2003, Sean Dockery wrote:
> Thanks, Tim, for the even handed response.
>
> I'm not looking for a business case to choose one or the other, however; it
> is certain that our customers will be deploying our application on both
> Linux and Windows (and even Solaris). I'm just looking to find out whether
> or not OS service (TCP/IP stacks, threads, file I/O, etc...) implementation
> differences between Linux and Windows have a significant impact on
> performance and thus should be weighed accordingly.
>
> I received a response in email from Peter Lin in which he details his
> experience (which was very helpful; thank you, Peter). I've read Peter's
> article about performance tuning and a few other white papers as well, but I
> haven't really seen anything in the past that focused on OS differences and
> how those differences might affect the recommended approach to profiling and
> tuning.
>
> My conclusions from my readings so far: Slow java code (i.e.: algorithms)
> will be slow on any platform; change the implementation to make it faster.
> Configurable behaviour dependent upon OS services (TCP/IP stacks, threads,
> file I/O, etc...) should be tuned for the platform on which the application
> will live.
>
> PS: I was sad to learn that the Tomcat Performance Handbook publishing date
> would be postponed. I would be thrilled if either you or Peter could tell
> me that the book will see a printer's press anytime soon.
>
> PPS: Is there a wiki for this stuff anywhere?
>
> "Tim Funk" <fu...@joedog.org> wrote in message
> news:3FDDADA1.9090403@joedog.org...
> > [I hate saying this since its rather very much like flambait but...]
> >
> > If its worth anything, I haven't had enough load on any of our apps to
> know
> > whether Linux or Windows is better. Instead, look at:
> > *** - Maintenance - If your a windows shop - stay windows ***
> > - Debugging - I think troubleshooting is easier on *nix systems (YMMV)
> > - Comfort - If your comfortable with unix concepts - linux might be easier
> > than windows
> >
> > -Tim
> >
> > Sean Dockery wrote:
> >
> > > I am planning to profile a web application on Windows XP (my development
> > > platform). I am curious as to whether or not different components in
> Tomcat
> > > and the JVM will behave differently (in a relative comparison) on Linux
> > > (production platform) than Windows.
> > >
> > > For example, I have had a person tell me that threads under Linux are
> more
> > > performant than threads under Windows--leading to the corollary that web
> > > applications under Linux are more performant than web applications under
> > > Windows on the same hardware. My guess is that this claim is based upon
> the
> > > supposition that thread/context switches under Linux are faster than
> under
> > > Windows. I find the claim rather dubious because I've never seen data
> to
> > > support the claim, but doubt is not certainty.
> > >
> > > Is there any evidence that this claim and other component performance
> > > differences between the Windows and Linux platform exist and are
> significant
> > > enough to throw my performance measurements out the "window". :-)
> > >
> > > My concern is that I'll profile the application under Windows and tune
> it,
> > > but then find that my gains aren't as significant or maybe even
> worthless
> > > under Linux.
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
>
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Tomcat performance on Windows versus Linux
Posted by Sean Dockery <us...@sbdconsultants.com>.
Thanks, Tim, for the even handed response.
I'm not looking for a business case to choose one or the other, however; it
is certain that our customers will be deploying our application on both
Linux and Windows (and even Solaris). I'm just looking to find out whether
or not OS service (TCP/IP stacks, threads, file I/O, etc...) implementation
differences between Linux and Windows have a significant impact on
performance and thus should be weighed accordingly.
I received a response in email from Peter Lin in which he details his
experience (which was very helpful; thank you, Peter). I've read Peter's
article about performance tuning and a few other white papers as well, but I
haven't really seen anything in the past that focused on OS differences and
how those differences might affect the recommended approach to profiling and
tuning.
My conclusions from my readings so far: Slow java code (i.e.: algorithms)
will be slow on any platform; change the implementation to make it faster.
Configurable behaviour dependent upon OS services (TCP/IP stacks, threads,
file I/O, etc...) should be tuned for the platform on which the application
will live.
PS: I was sad to learn that the Tomcat Performance Handbook publishing date
would be postponed. I would be thrilled if either you or Peter could tell
me that the book will see a printer's press anytime soon.
PPS: Is there a wiki for this stuff anywhere?
"Tim Funk" <fu...@joedog.org> wrote in message
news:3FDDADA1.9090403@joedog.org...
> [I hate saying this since its rather very much like flambait but...]
>
> If its worth anything, I haven't had enough load on any of our apps to
know
> whether Linux or Windows is better. Instead, look at:
> *** - Maintenance - If your a windows shop - stay windows ***
> - Debugging - I think troubleshooting is easier on *nix systems (YMMV)
> - Comfort - If your comfortable with unix concepts - linux might be easier
> than windows
>
> -Tim
>
> Sean Dockery wrote:
>
> > I am planning to profile a web application on Windows XP (my development
> > platform). I am curious as to whether or not different components in
Tomcat
> > and the JVM will behave differently (in a relative comparison) on Linux
> > (production platform) than Windows.
> >
> > For example, I have had a person tell me that threads under Linux are
more
> > performant than threads under Windows--leading to the corollary that web
> > applications under Linux are more performant than web applications under
> > Windows on the same hardware. My guess is that this claim is based upon
the
> > supposition that thread/context switches under Linux are faster than
under
> > Windows. I find the claim rather dubious because I've never seen data
to
> > support the claim, but doubt is not certainty.
> >
> > Is there any evidence that this claim and other component performance
> > differences between the Windows and Linux platform exist and are
significant
> > enough to throw my performance measurements out the "window". :-)
> >
> > My concern is that I'll profile the application under Windows and tune
it,
> > but then find that my gains aren't as significant or maybe even
worthless
> > under Linux.
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org
Re: Tomcat performance on Windows versus Linux
Posted by Tim Funk <fu...@joedog.org>.
[I hate saying this since its rather very much like flambait but...]
If its worth anything, I haven't had enough load on any of our apps to know
whether Linux or Windows is better. Instead, look at:
*** - Maintenance - If your a windows shop - stay windows ***
- Debugging - I think troubleshooting is easier on *nix systems (YMMV)
- Comfort - If your comfortable with unix concepts - linux might be easier
than windows
-Tim
Sean Dockery wrote:
> I am planning to profile a web application on Windows XP (my development
> platform). I am curious as to whether or not different components in Tomcat
> and the JVM will behave differently (in a relative comparison) on Linux
> (production platform) than Windows.
>
> For example, I have had a person tell me that threads under Linux are more
> performant than threads under Windows--leading to the corollary that web
> applications under Linux are more performant than web applications under
> Windows on the same hardware. My guess is that this claim is based upon the
> supposition that thread/context switches under Linux are faster than under
> Windows. I find the claim rather dubious because I've never seen data to
> support the claim, but doubt is not certainty.
>
> Is there any evidence that this claim and other component performance
> differences between the Windows and Linux platform exist and are significant
> enough to throw my performance measurements out the "window". :-)
>
> My concern is that I'll profile the application under Windows and tune it,
> but then find that my gains aren't as significant or maybe even worthless
> under Linux.
---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-user-help@jakarta.apache.org