Zimbra webmail SLOW

Discuss your pilot or production implementation with other Zimbra admins or our engineers.
mafiabusiness
Advanced member
Advanced member
Posts: 53
Joined: Sat Sep 13, 2014 3:28 am

Re: Zimbra webmail SLOW

Post by mafiabusiness »

Also, I asked you some other things in the previous comment that you still not answered...

— apologies…

- How big is your mariadb database and how much memory did you assigned to the innodb buffer?
- Which parameters do your java process execute with?
- Chech for abnormal garbage collection times for both new gen and long term memory areas.
- Can you see any deny of service issue like jetty DDoS filters being applied with a delay for new petitions?

— I will ask them to see what they say.

I have had a look around in my.conf and found that
- max_connections = 300
innodb_buffer_pool_size = 20Gb
innodb_buffer_pool_instances = 18
innodb_open_files = 6000

…and now for example, if I run the "free -m" command, I get

free: 733
+/- buffers-cache: 64690
Swap: 16039 (vm swap is off, used is 0)

Miguel
User avatar
pup_seba
Outstanding Member
Outstanding Member
Posts: 687
Joined: Sat Sep 13, 2014 2:43 am
Location: Tarragona - Spain
Contact:

Re: Zimbra webmail SLOW

Post by pup_seba »

Hi,

Things I notice by your last post:
- You have over 60Gb of available ram to be used at anytime.
- You don't seem to have excesive ios.

Maybe...
- The noatime is not configured in your fstab (explained here https://wiki.zimbra.com/wiki/Performanc ... eployments)
- zmlocalconfig mailboxd_java_heap_size would return a value of 6144 (see here about values that should be used for a system with 100GB system like yours...more about 20GB and not 6GB as it seems to be your case). But changing this value should be done carefully, as such a huge amount of memory for a jvm, may lead to excesive garbage collection times making your zimbra even slower.
- The innodb_max_dirty_pages_pct is configured to 30 instead of 10 (which could make your iops spike from time to time instead of sustaining a more consistent iops over time).
- innodb_buffer_pool_size undersized (explained here https://wiki.zimbra.com/wiki/Performanc ... eployments) and the database (under /opt/zimbra/db/data) quite bigger than the value for your buffer_pool.
- You still did not answer about the ddos filters being fired, but just in case check this https://wiki.zimbra.com/wiki/DoSFilter.

All of this could help, but if your problem is slow disk or even if it's not, you'll see things get better by moving to faster disks.

I'll drop this thread now.

Gl!
User avatar
pup_seba
Outstanding Member
Outstanding Member
Posts: 687
Joined: Sat Sep 13, 2014 2:43 am
Location: Tarragona - Spain
Contact:

Re: Zimbra webmail SLOW

Post by pup_seba »

Oh...I just saw the last message of yours. Ok. So you are on to db tunning :)

Again, I wish you good luck. Not much help I can provide now.
Post Reply