users@glassfish.java.net

Re: Performance T1000 vs Intel

From: Dick Davies <rasputnik_at_hellooperator.net>
Date: Fri, 9 Jan 2009 12:06:14 +0000

On Thu, Jan 8, 2009 at 2:48 PM, <glassfish_at_javadesktop.org> wrote:
> So I got my T1000 (Cool threads) in and I was all giddy! Especially that we are an all MS shop with Intel.
>
> How I got to this point... We have a "web service" (XML over HTTP POST) running as an ISAPI DLL on IIS. As a customer requirement we have to create a WS* version. So I suggested we use Metro and Glassfish.
>
> ISAPI DLL: Accept HTTP POST XML, reply XML "Hello World!"
> Metro/Glassfish: Accept SOAP, reply SOAP "Hello World!"
>
> The numbers where astonishing:
> ISAPI DLL on 4 core 2.x GHz Intel IIS 6: 1000 requests a second.
> Metro/Glassfish: 4 core 2.x Ghz Intel: 3000 requests a second.
>
> Apparently MS claims that IIS 7 is 70% faster, but that still doesn't come close to Glassfish. Which then prompted us to see if we can get even more using a Sparc and SUN technology.
>
> So here we are now...
> T1000 Sparc, 1.0 Ghz, 6 core, 8 Gigs RAM, Solaris 10 11/06 stock except installed JDK 1.6_11 and App Server 9.1
> Intel 2.x Ghz, 4 Core, 1 Gig RAM, Windows 2003 latest patches, JDK 1.6_10, Glassfish 2ur2
>
> Both boxes configured as per Scot Oaks blog: http://weblogs.java.net/blog/sdo/archive/2007/12/a_glassfish_tun.html

If it's a T1000 make sure you read the CMT sections of that post.
>
> Now on each machine I'm getting 3000 requests per second. Do you guys think that the T1000 can do better and what can I do to get the most out this box?

You have a silly amount of CPU cores in a T1000. I'd expect you to do
very well at high concurrency levels.
have you done similar benchmarks and increased the number of
simultaneous client connections?