It might be Nagle's algorithm on one of the sockets involved. A simple
test with snoop should show precisely where the delay is coming from.
-tom
-----Original Message-----
From: Douglas E. Hornig [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, May 02, 2001 3:42 AM
To: [EMAIL PROTECTED]
Subject: RE: Solaris Sparc Performance Problem
I ran truss and it generated a ton of stuff that I frankly had trouble
making much sense of. My approach for now is to work with some Sun
engineers to try to get to the bottom of it. So far they have suggested
trying the latest JDK (J2SE 1.3.1 RC2), and passing the requests through
apache (I'm running tomcat standalone now). The latter may make the problem
go away since apache doesn't seem to suffer from the problem, but I consider
it avoiding the problem rather than solving it. I have not tried either
yet.
I'll keep you all posted if anything turns up.
Doug
--- You wrote:
Maybe I missed the response, but whatever became of this issue?
Thanks,
Dave
-----Original Message-----
From: Arieh Markel [mailto:[EMAIL PROTECTED]]
Sent: Tuesday, April 24, 2001 10:31 AM
To: [EMAIL PROTECTED]
Subject: Re: Solaris Sparc Performance Problem
Here is what I would do to see the differences:
. on Solaris, run the application through truss
. on Linux, run the application through strace
This should yield information about where the time is being spent.
I am also wondering whether the Solaris machine is properly configured
with regards to things like nameserver lookups, proxy setups, etc.
Arieh
> Mailing-List: contact [EMAIL PROTECTED]; run by ezmlm
> Delivered-To: mailing list [EMAIL PROTECTED]
> From: [EMAIL PROTECTED] (Douglas E. Hornig)
> Subject: Solaris Sparc Performance Problem
> To: [EMAIL PROTECTED]
> Content-Disposition: inline
> X-MIME-Autoconverted: from quoted-printable to 8bit by
amon.Central.Sun.COM id
LAA19808
>
> I posed this problem to the good folks on the users list. While they are
a
great bunch, and several offered some suggestions, I was unable to get any
help
from them so I'm trying the dev list now.
>
> The problem in a nutshell is that requests I make to tomcat running on a
Solaris Sparc from a Windows client take at least 0.15 to 0.20 seconds. If
I
run tomcat on a Linux PC, or use a Linux PC as a client instead of Windows,
the
turnaround time is more like 0.01 seconds.
>
> Here are the particulars:
>
> * All machines are on the same 100Mbit ethernet.
> * Tomcat is running standalone.
> * I tried a couple of different Sparcs, a 420R and an Ultra 5, neither
with
any load. No difference.
> * I wrote a simple Java program to use as the test client so there are no
browsers involved.
> * I tried various different Java VM releases on the Sparcs, 1.2.1 and
1.3.0
with no difference seen.
> * I tried a couple different PCs (NT4 and Win2000) and found the same
results.
> * Other programmers here reported slowness using VisualBasic as the client
instead of Java (that's how I got started investigating this). Java Web
Server
2.0 also appeared to have the same problem as tomcat. I have not personally
been able to verify these assertions.
> * The results seem very repeatable.
> * I used a generic tomcat 3.2.1 for the server and hit the
examples/servlet/HelloWorldExample URL for these tests.
>
> This is a very serious problem for us. The above mentioned VB client that
we're developing can make dozens of calls to the server per screen, so those
0.2
second delays add up. I like Linux a lot myself but the bosses here feel
more
comfortable with more traditional business models, and besides shouldn't
Java
run best on a Sparc with Solaris? I am perplexed as to what the problem is
and
would greatly appreciate any help or ideas I can get.
>
> Thanks in advance,
>
> Douglas Hornig
> Dartmouth-Hitchcock Medical Center
> Lebanon, NH
--
Arieh Markel Sun Microsystems Inc.
Network Storage 500 Eldorado Blvd. MS UBRM11-194
e-mail: [EMAIL PROTECTED] Broomfield, CO 80021
Pray for snow !!!! Phone: (303) 272-8547 x78547
(e-mail me with subject S
--- end of quote ---
