So... we are about to torpedo our VDI deployment because it is just not performing to expectations. I am hoping someone can point out what we might have overlooked.
Cluster:
4 Dell R900 servers with 4 Xeon X7350 2.9Ghz CPU's (4 core) and 128GB RAM
Total Cluster Resources: 64 pCPU's and 512GB Memory
Datastore:
EMC Clarrion attached via 2 port QLogic 4Gb HBA on each ESXi host
EMC Celerra for CIFS
Network:
4 1GB adaptors on each ESXi host, 2 for mgmt/vmotion, 2 for VM traffic (CIFS, agent, web, etc)
Everything is connected via 1GB ethernet
VMware:
ESXi v5.0
vCenter v5.0
View 5.1.1
Workstation Setup:
Teradici zero client, Firmware 4.0
Dual monitors
2 vCPU
3 GB Memory
Windows 7 32-bit
Roaming profiles and folder redirections to CIFS share
Desktop Pool Setup:
Replica disks stored on SSD
Linked clones stored on tiered LUN (15K FC, 10K FC, 7.2K SATA)
Floating Pool
Host Cache enabled (2GB)
With the above equipment, we are lucky to get 50 users (our target is 350) on the system before it completely bogs down and everyone is complainging. We have graphed our network and there is no indication of a bandwidth bottleneck. Plenty of pipe and all the interface error counters are zero. Our average zero client (PCoIP) bandwidth usage for 60 desktops is only around 30Mbps, peak is about 44Mbps. Our Celerra and Clarrion show only about 30% utilization on the storage processors. We are barely getting a vCPU/pCPU ratio of 2:1 when I hear even power users shoud have ratios closer to 6:1 and standard desktop users more like 15:1.
Our equipment is kind of old, but is that really the problem? Can Win7 desktops be this intensive? Our users are mostly on the web and use a handful of java apps.
This resource graph shows our cluster at 64 desktop users. It may look like there is some room, but the user experience doesnt feel like it. We are hoping our demo of VCOps and VCOps for View will help, but nothing so far.