I'm getting ready to slowly migrate about 2000 users over to Zimbra. I am currently preparing a RHEL 4.5 server on a Xen virtual machine with storage located on a SAN. So here are the details:

Server Physical Hardware:

-Two Dual Core Xeon 64-bit CPUs (Intel(R) Xeon(R) CPU E5335 @ 2.00GHz)
-8 gigs of RAM
-36 gig internal drive (mirrored with two 36 gig drives)
-Running CentOS 5 with Xen virtualization enabled only using 512 megs of RAM
-Dual fiberchannel cards to connect to SAN
-516 gigs of disk space carved out of SAN and presented to CentOS as /dev/Zimbra/virt-hda LVM logical volume

Virtual Server:

-Xen, using HVM instead of paravirtualization, configured to use 4 gigs of RAM, with 4 VCPUs (might change this to 2 VCPUs since the dual core nature of the real CPUs doesn't seem to translate)
-/dev/Zimbra/virt-hda on the "Host" system is presented as a 516 gig /dev/hda in the "guest" system and is allocated as 100 megs boot, about 500 gigs for the root filesystem and 16 gigs for swap

I really want to test this very thoroughly before I commit to actually running this way and the best place to test the load would be at the application (Zimbra) level. Does anyone know of any way to simulate 2000 users hitting the box via the Zimbra web mail application as well as about 250 IMAP users with mail stores anywhere from 10 megs to 200 megs. They'd be checking mail, on average about once every five minutes. My main concern is I/O performance since I've heard that both network and disk I/O may be issues when running in an HVM "guest" in Xen. But I really don't know how that will interact with Zimbra in a real world situation. Any suggestions?