poor performance ZCS NE vs. ZCS OSS

[fisch09]

Hi all,

I am concerned about the CPU load on my ZCS NE installation.
I have ZCS 4.5.6 NE and ZCS 4.5.6 OSS running on identical hardware.

CPU: Athlon 64 Processor 3700+
Memory: 1GB
Hard drives: 2x 160GB sata (soft raid 1)
OS: Ubuntu server 6.06 (32bit)

both servers have 20 users and about 2GB data in /opt/zimbra/store

ZCS OSS is running fine and the CPU load never goes over 0.6 but on ZCS NE the CPU load goes up to 8 in short intervals (3x or 4x an hour)

I already disabled HTML attachment view and attachment indexing on ZCS NE as I suspect one of the additional features in NE to cause the problem.

Do you have any suggestions where I could start to dig ????

Cheers
Andre

[jholder]

Hi Andre-
Sorry you're having trouble.

It is odd that there is such a large performance degradation problem between the Open Source version and Network Edition version.

One option would be to kill java (kill -3) and grab a thread dump in the catalina.out file and see what it's doing.

Does "top" reveal any useful information?

[fisch09]

here the top

top - 08:32:21 up 48 min,  1 user,  load average: 7.15, 7.20, 5.75
Tasks:  94 total,   2 running,  92 sleeping,   0 stopped,   0 zombie
Cpu(s): 100.0% us,  0.0% sy,  0.0% ni,  0.0% id,  0.0% wa,  0.0% hi,  0.0% si
Mem:   1034656k total,   991080k used,    43576k free,    15512k buffers
Swap:  2096376k total,       96k used,  2096280k free,   192744k cached

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND            
 4716 zimbra    18   0  736m 352m  13m S 99.7 34.9  35:11.26 java

at this time there have been 2 users connected an NO emails in any Q

[code]
4716 ? Sl 37:10 \_ /opt/zimbra/jdk1.5.0_08/bin/java -Xms404m -Xmx404m -client -XX:NewRatio=2 -Djava.awt.headless=true -Dcatalina.base=/opt/zimbra/apache-tomcat-5.5.15 -Dcatalina.home=/opt/zimbra/apache-tomcat-5.5.15 -Djava.io.tmpdir=/opt/zimbra/apache-tomcat-5.5.15/temp -Djava.library.path=/opt/zimbra/lib -Djava.endorsed.dirs=/opt/zimbra/apache-tomcat-5.5.15/common/endorsed -classpath /opt/zimbra/apache-tomcat-5.5.15/bin/bootstrap.jar:/opt/zimbra/apache-tomcat-5.5.15/bin/commons-logging-api.jar:/opt/zimbra/lib/jars/zimbra-launcher.jar com.zimbra.cs.launcher.TomcatLauncher
[\code]

catalina.out is about 6000 lines, should I post it here?

Kind Regards
Andre

[fisch09]

Hi,

I found an old thread here in the forum about high java resource usage and MAC imap client. I do not know if this important but we have MAC imap clients.

Cheers
Andre

[jholder]

By looking at your catalina.out, looks like you have several Zimbra Mobile Clients going? Is that right?

I think that might be part of why java is higher than the OSS version.

[fisch09]

Hi all,

it took a while but I found the reason for the high CPU load and I can reproduce it. If an email is stuck in the deferred queue the CPU load sky rockets (up to 20 on my system)

What happened exactly was:

I had no catch all accounts defined and when an email hits the system for an undefined account it bounces back. Zimbra then notifies the sender that the email could not be delivered. If the sender address is faked and zimbra can't deliver the "could not deliver" email its stuck in the deferred queue - by default for 5 days.
Every single time I had one or more emails in the deferred queue the CPU load goes up - delete the email from the Q and everything is fine again.
After I assigned a catch all for every domain the CPU load never goes above 1.0 and averages below 0.5.

Does anyone has similar issues when mails are stuck in the deferred queue?
I attached a png that shows the cpu load for the last days.

Cheers
Andre