During the CeBit 2010 we managed to reach a number of extremes. Not only were we able to get vCenter to reach new heights, we also pushed the envelope on system densities. Together with our Dutch reseller ForceFusion we managed to setup a full operational system in a little over a day.

The hardware is pretty much overkill, as there is a VMax with 100TB FC and 2TB of EFD there, as well as a whiptail and a greenbytes storage system. A Dell blade center with 11 current dual and quad socket quad core machines and nice big memory, the remaining blades were used for vCenter, MSSQL, VeeAm and others. All of it tied together using an Arista 7148SX with 960Gbit/s backplane speed and low latency, 10gig sfps. (Thanks EMC, Intel, Consolidate-IT and ForceFusion for the hardware)

The technology shows some interesting behaviour, as it seems the overall solution is getting faster the more desktops we deploy. This is expected behaviour for larger VirtualStorm environments.

The bottlenecks right now are mostly VirtualCenter, Tomcat and MSSQL. Optimization scripts have allowed vCenter to go up to 4500 machines. Some of the larger blades are pushed to the limits, with one blade pushed to 826 VMs. (See the pictures below)

Depending on workload the switch pushed between 100 and 250 Gbit/s through the entire system

The desktops that were deployed together resulted in an application stack of a maximum of 1125 TB of SWV managed applications. We had 60GB of apps uploaded, so the actual application deployment could reach 280 TB on 4500 VMs.

Access to desktops occured dinamically through one of the available Panologic devices.

Deployment time for 150 desktops: 3 minutes. (Intel Nehalem sure does the job) Depending on load on the overal system deployment times for desktops ranged from 0.5 to 2 seconds. Delays were mostly caused by registering of the VMs in vCenter; that takes 70% of the overall duration.