Virtualizing Exchange Part II

A few weeks ago I wrote about a virtualized Microsoft Exchange solution that EMC published under Microsoft's Tested Exchange Solutions Program (ETS) in which we tested and documented a completely virtualized Exchange 2010 configuration for 30,000 mailboxes on EMC Unified Storage.

Recently we released our second Tested Exchange Solution for Exchange 2010 and this time we teamed up with Brocade and Dell to document how we virtualized Exchange using Microsoft Hyper-V. The paper entitled Zero Data Loss Disaster Recovery for Exchange 2010, includes EMC's implementation of a Microsoft Third Party Replication API called EMC Replication Enabler for Exchange (REE) using EMC's Mirrorview for array based synchronous replication.

Side note: As an alternative to Exchange's native DAG replication, EMC developed REE as a free software utility. REE uses block-level synchronous replication over Fibre Channel (opposed to Exchange DAG which is network based asynchronous). REE integrates with the Microsoft DAG third-party replication API to enable local storage as well as synchronously replicated storage.

Some highlights of this published white paper include:

• Exchange Server 2010 on Microsoft Hyper-V R2 with 10,000 users per site across two sites.
• 20,000 total users with 500MB mailboxes, 150 messages per day and two Exchange DAGs utilizing EMC's REE 3^rd party replication API.
• Two EMC CLARiiON CX4-480's (1 per site) with 80 15k rpm drives per location.
• 8 Exchange mailbox VMs with 2500 users per server and 10 disks per VM.
• EMC Replication Manager w/Snapview for VSS based application consistent replicas for local protection.
• Brocade ServerIron ADX application load balancing.
• Brocade SAN and IP switching.
• Dell PowerEdge R910 servers.
• Testing was completed in the Microsoft EEC labs in Redmond, WA.

 Key findings from the tests include:

• Exceeded performance expectations using Microsoft Jetstress test tool for 20,000 users with read and write latencies below Microsoft's recommended threshold for Exchange including with and without synchronization of data.
• Validated the virtualized Exchange environment could run under normal operating condition during a peak load. During peak loads the average CPU utilization of the mailbox VMs was approximately 26%.
• Demonstrated the ability to failover to a stand-by site and failback without any impact to the performance of the Exchange mail environment.
• EMC's synchronous based REE enabled the ability to achieve zero data loss with the lowest possible RPO.
• Brocade's ServerIron ADX provided affinity and load balancing for the Exchange traffic between servers.
• Achieved 3:1 consolidation using Hyper-V virtualization by consolidating 20,000 users onto four Dell PowerEdge 910 servers between two datacenters. In a physical environment this would have required 32 servers.

In the end, performance achieved was much better than the designed targeted baseline. A single CX4-480 achieved 8,120 Exchange 2010 transactional IOPS across eight Exchange virtual machines (or 2,288 IOPS over the designed solution). This additional headroom provides additional buffer in the event of an increase in mail activity or as the environment grows over time. Below are the published Jetstress results for Exchange 2010 on the CX4-480:

Like the previous virtualized Exchange solution, this continues to reinforce that Microsoft Exchange 2010 is a good candidate for virtualization and that Hyper-V easily supports enterprise workloads like this one.

For more information on this solution as well as EMC's Replication Enabler for Exchange (REE), I encourage you to check out Dustin Smith's excellent blog and write-up on this solution.