Slow Sending Messages with Zimbra?

[Diranged]

The problem: I have a few users complaining almost daily that sending messages is extremely slow with their IMAP clients. They will hit send sometimes and the message will take between 20-seconds and forever (aka, it hangs) to send some messages. Other times the messages go out in just a few seconds. This seems to be much worse with Thunderbird, but still happens with the Apple Mail.app client.

Here's a bit of background...

We recently switched from a hand-built pair of linux-HA'd mail VM images (dovecot, postfix, amavisd, spamassasin) storing active mail queue's on VMFS disks (over NFS from netapp filer). user maildirs were mounted via nfs. With this setup I never had any performance complaints, and average vmware server load was around .8-1.2 with 2 vcpus and a gig of ram. all accounts were managed via ldap.

Our new setup is a single VM image (4 vcpus, 4GB ram, VMFS volume mounted from a NetApp via NFS). Running the latest Zimbra OpenSource edition, all services on this single image. zimbra ldap, amavis, spamassassin, web ui, etc all running. Currently the Zimbra filesystem is a separate VMFS disk image (again, vmware mounts these via nfs) running ext3. The zimbra message store is CURRENTLY a direct netapp nfs mount with about 35GB of messages.

Generally the vm image seems to have a load of between .4 and 1.1 depending on how many messages are coming in at any given moment... we have ~50 users and receive 20-30k real messages per day (not including spam thats blocked before it gets into the system). Only 2-3 users use the web interface, the vast majority use IMAP clients. All of the IMAP clients are using IMAP SSL (993) and we are using SSL on ports 465 and 587 for authenticated outbound SMTP.

*** i am aware that NFS is a slow mailstore back end. I have built an iSCSI lun and synced the messages over, but I need to wait until tonight to do the switchover in Zimbra. ***

I really could use some recommendations on how to troubleshoot this... I just don't really see these slow mail sends myself, and neither do most of our users. The two users who are complaining the most each have the most amount of mail in their accounts (3.5GB and 5GB), but there are quite a few other users who are above 2GB.

[Diranged]

I've made a number of changes tonight, and will see how my users react.

1) We've moved both the message store as well as the actual zimbra install onto an iSCSI volume on our NetApp filer.
2) The VM image has been set to use the deadline elevator
3) Adjusted clamav.conf.in to set the tempdir to /tmp, which is a 256MB tmpfs volume rather than using the default location.

[pheromone]

Hello Diranged,

Were you able to figure out why sending was slow?

We are experiencing similar symptoms:

- when a user sends a mail internally using Thunderbird, the mail is sent quickly, because I don't think that the server authenticates client.

- if the mail is sent to an external domain, it takes a lot of time to authenticate (10-20 seconds) and then sends the message.

Daniel
Pheromone

[dalmate]

@Diranged: did you check how many imap connections run when your imap service slow?
netstat -ant| grep 143 | wc -l
with zimbra user
Maybe you should increase the number of imap concurent connections on your server. I used 500 concurrent connections and it run well.

[pheromone]

It turns out, we demoted and decommissionned our AD server that was first in the list of servers for LDAPS authentication in Zimbra configuration.

Therefore, when a user would send an e-mail, authentication would take some time (10-20 seconds), timeout, and then use the second LDAP server in the list to send the message successfully.

Hope this can help someone.

Regards,

Daniel

[bartounet16]

Yes, i have notice too, that a bad DNS resolution can slowing down mail delivery

For example if you have 2 DNS server setted in your configuration, if the first does not responding, zimbra wait the time out to use the second, so mails delivery will be slowed down