We've just starting out with moving from zimbra OS to NE and in the processes of migrating over our first two accounts I've noticed a dramatic speed difference which I can't explain.
We started out testing zimbra on a centos vmware server: Release 5.0.18_GA_3011.RHEL5_20090707164432 CentOS5 FOSS edition. This server has a single CPU and only 2 gigs of ram. The mail store/index are on an iscsi mount on our SAN formated as ext3 in an LVM.
Our new production server (Release 5.0.18_GA_3011.RHEL5_64_20090707153751 RHEL5_64 NETWORK edition) has 8 cores and 16 gigs of ram and *isn't* a vmware server. The mail store/index for this one is also an iscsi mount from the same SAN formated as ext3 but not in an LVM.
What surprised me is that when I export an account and copy it over to new NE server and import it its taking twice as long to import as it does on the older open source vmware server. On the new NE server running the command:
zmmailbox -z -m email@example.com postRestURL '//?fmt=tgz' /tmp/user1.tgz
takes ~60 minutes
The same command run on the OS vmware server takes ~30 minutes.
The account being imported is 202M compressed and 368M once imported.
Watching resources on the new NE server it doesn't look like IO or is being hit very hard. CPU is ~80% idle ~15% IO wait. Watching "iostat -m 1" shows the iscsi never comes close to its max speed.
I've looked at and tried a few options out of Performance Tuning Guidelines for Large Deployments - Zimbra :: Wiki but it hasn't changed much.
After seeing this I was worried about continuing the migration of users over to the new server. Any suggestions on what else I can try? Is importing a user with postRestURL a poor performance measurement?