[Deploying Sakai] [Building Sakai] Solaris Architecture Affects Performance?

[mizematr at notes.udayton.edu]

The JVM garbage collection settings we use were actually the second thing 
we started looking at (after mucking around in the code for a little 
while).  When we had first installed Sakai GC was responsible for some 
loooooooong pauses as it was running.  We have since switched to the 
concurrent low pause collector.  This got rid of the long pauses that we 
saw but the overall performance of our server was still pretty slow.


Our current JVM settings:
JAVA_OPTS="-server \
-d64 \
-XX:+UseConcMarkSweepGC \
-XX:+UseParNewGC \
-XX:+CMSParallelRemarkEnabled \
-Djava.awt.headless=true \
-Xms6144m -Xmx6144m \
-XX:NewSize=1024m \
-XX:MaxNewSize=2048m \
-XX:PermSize=512m -XX:MaxPermSize=512m \
-Xloggc:/opt/sakai/gc_log/gc.log -XX:+PrintGCDetails 
-XX:+PrintGCApplicationStoppedTime -XX:+PrintGCApplicationConcurrentTime 
-XX:+PrintGCTimeStamps -XX:+PrintTenuringDistribution \
-XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath=/opt/sakai/gc_log \
-Dcom.sun.management.jmxremote"
---------------------------------------------------------------------------
Matt Mize, Systems Administrator
 - Pay no attention to the man in the back office....
Matt.Mize at notes.udayton.edu
(937) 229-1024

UDit Department, University of Dayton
300 College Park, Dayton, OH, 45469-1302



Branden Visser <branden at uwindsor.ca> 
01/27/10 05:08 PM
Please respond to
branden at uwindsor.ca


To
mizematr at notes.udayton.edu
cc
sakai-dev at collab.sakaiproject.org, production at collab.sakaiproject.org
Subject
Re: [Building Sakai] Solaris Architecture Affects Performance?






mizematr at notes.udayton.edu wrote:
> 
> We recently completed a year long process to migrate our users off of 
> WebCT CE 4.1 and onto Sakai 2.6.1.  The transition went well and for the 

> most part, our Faculty and Students are very happy with Sakai, but there 

> has been one complaint:  "Sakai seems slower than WebCT."  At first we 
> didn't take this too seriously, figuring it was a relative thing and 
> users would soon become used to it, but as we looked into it, we found 
> that our instance of Sakai was not just slower than our instance of 
> WebCT, it was slower than instances of Sakai at other schools in the 
> community.  We've been working on this problem for the past several 
> months, and believe we have finally found a major part of the problem, 
> the Solaris architecture.
> 
> When we bought a box for our Sakai installation, we bought one of the 
> latest and greatest Solaris machines at the time, a Sun Blade T6300 and 
> we think this is the problem.  One of our Solaris admins found a report 
> that some applications cannot leverage the architecture of the T6300 and 

> will in fact run slower on the new architecture over the older 
> architecture.  We setup two test machines on the older architectures and 

> repeatedly loaded a gradebook with 50 students and 44 columns, with 
> these findings:
> 

In the past, our appserver cluster has been a split between T2000's and 
a T5120. I didn't run benchmarks to pit them against each other, but our 
load-balancer consistently favoured the tomcat instances on the T5120, 
which is a sign that it is yielding a quicker response. Besides, they're 
both sun4v, so I guess it's a bit irrelevant.

Did you happen to spit out the GC work a-la:

-verbose:gc -XX:+PrintGCTimeStamps -XX:+PrintGCDetails 
-XX:+PrintTenuringDistribution -Xloggc:$CATALINA_HOME/logs/gclog.out

? This could give some leads if the garbage collectors are simply 
reacting differently and causing more pauses.

Branden

> Hostname:         sakaitest
> Database:        Oracle, sktest
> uname -m:        sun4v
> Median Speed:        4.46s
> 
> Hostname:        kaylee
> Database:        Oracle, sktest
> uname -m:        sun4u
> Median Speed:        2.71s
> 
> Hostname:        river
> Database:        Oracle, sktest
> uname -m:        i86pc
> Median Speed:        1.50s
> 
> So, my questions are:
> 
> Has anyone else seen behavior like this?
> Are we just missing a setting for Java on the sun4v chipset?
> Solaris schools, what kind of architecture are you running?
> 
> Any information/thoughts would be much appreciated.
> 
---------------------------------------------------------------------------
> Matt Mize, Systems Administrator
> - Pay no attention to the man in the back office....
> Matt.Mize at notes.udayton.edu
> (937) 229-1024
> 
> UDit Department, University of Dayton
> 300 College Park, Dayton, OH, 45469-1302
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> sakai-dev mailing list
> sakai-dev at collab.sakaiproject.org
> http://collab.sakaiproject.org/mailman/listinfo/sakai-dev
> 
> TO UNSUBSCRIBE: send email to 
sakai-dev-unsubscribe at collab.sakaiproject.org with a subject of 
"unsubscribe"

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://collab.sakaiproject.org/pipermail/production/attachments/20100128/946b6e8c/attachment.html