Jim Sanzone, one of my vSpecialist colleagues and VDI master was joined by the VCE team to demonstrate how two Vblocks with VPLEX within a datacenter can provide a **highly** available “Always ON” infrastructure for VDI environments that require lavish SLA’s.
EMC VPLEX Local provides seamless data mobility and high availability within a data center. It also allows you to manage multiple heterogeneous arrays from a single interface.
The idea behind EMC VPLEX Local for VDI is that in the event of a total storage array failure all virtual desktops would still run without service interruption. In order to do that, VPLEX synchronizes storage IOs across different arrays.
There are other ways to achieve greater levels of availability for VDI deployments and I have discussed them in my article VMware View Disaster Recovery Scenarios & Options. However, these solutions would only benefit Floating (View) and Pooled (XenDesktop) desktops. They do not protect against total storage failure in a single datacenter.
If your VDI implementation is targeting cost reduction this may not be the right a solution for your organization. Organization like hospitals, or industries requiring highly available VDI platforms may suit this VPLEX setup.
The VDI solution used was Citrix XenDesktop 5 running PVS (Provisioning Services), Windows 7 Enterprise, MS Office 2010, LoginVSI Workload Simulator and Liquidware Labs Stratusphere UX.
Vblocks were configured with Cisco UCS B200 M2 blades with 96GB. For storage the 1st Vblock had a EMC VNX5700 array and the 2nd Vblock had a EMC VNX7500 array.
The test had four objectives:
- Demonstrate basic performance attributes of XenDesktop in a Multi-Vblock/VPLEX design
- Demonstrate High Availability features of the Vblock
- Demonstrate Operational aspects of managing the Vblock + VPLEX combination
- Demonstrate the Ultra High Availability of a Vblock + VPLEX design
The picture below demonstrate the Vblocks distribution with UCS Chasis and the VNX arrays.
The picture below demonstrate the VPLEX/VNX configuration.
Below is a list of the simulated failures executed at the infrastructure layer while trying to disrupt services. The team was not able to disrupt services even when disconnecting the Fabric Interconnects or MDS Storage switches. The only times virtual desktops went offline were during the HA event when they were being restarted at the secondary Vblock at a total Vblock simulated catastrophic failure. Really cool!
I know Jim is working on a detailed article about his tests and I’ll publish the link here when it’s available.