Case Study: Exchange 2007 on VMware and NetApp
There are a lot of tangible benefits to running Exchange in a virtualized environment rather than a physical one, but while a lot of people are evaluating such solutions, it seems that many still remain skeptical about performance and reliability.
Migration to VI3 and Exchange 2007
A year and a half ago, we had roughly 300 mailboxes on Exchange 2003. Since then we’ve grown to about 450 mailboxes (seats). While that’s not big in absolute numbers, it represents 50% growth in our Exchange environment (and other associated infrastructure) over a relatively short time frame.
Table 1) Comparison of Exchange 2003 versus 2007.
At this point, we started by doing our homework; we watched Webcasts, read books, and so on until we were comfortable with Exchange 2007. Next we created a few test mailboxes in Exchange 2003 and installed an Exchange 2007 server in our virtual environment. We then migrated the test mailboxes from Exchange 2003 to Exchange 2007 and tested the send and receive functions both internally and externally.
Advantages of the New Environment
VMotion allows us to migrate virtual machines (VMs) from one server to another without interrupting service. For example, if I need to do a BIOS update or other firmware update on a server, I can migrate all the VMs off of that server and take the server down for the upgrade without the users even noticing that anything has happened.
Figure 1) VMotion.
VMware DRS gives us automatic load balancing across our ESX servers so that we get optimal utilization such that a server is never overloaded or underutilized. VMware HA protects us by automatically restarting VMs on other servers should an entire ESX server fail.
Figure 2) VMware DRS uses VMotion to automatically migrate VMs to balance load. VMware HA automatically restarts VMs on another server if a physical server fails.
One unique thing we’ve done in our environment is to create a rule that keeps the VMs that run our Exchange 2007 mailbox server, our hub and client access server, our Symantec™ Enterprise Vault™ Archive server, and our BlackBerry server on the same ESX server. That way all the communication that occurs between those servers actually happens on the virtual switch within the ESX server and doesn’t traverse our physical network switches.
Figure 3) Traffic through network switches
The best practices for Microsoft® Exchange suggest using RAID 10 (striping and mirroring) for back-end storage. However, mirroring—which requires double the storage capacity—was an expensive option for us. NetApp gave us the ability to use RAID-DP® to provide the high I/O performance required by Exchange at much less cost without sacrificing resiliency. In fact, NetApp’s Double Parity RAID implementation actually provides better resiliency than RAID 10. RAID-DP is 163 times less likely to lose data than RAID 10 and 4,000 times less likely to lose data than RAID 5.
Table 2) Comparison of RAID 5, RAID 10, and NetApp RAID-DP.
Our ESX servers have Fibre Channel HBAs, which connect them to our NetApp storage system. While we’re using Fibre Channel today, one advantage of the NetApp approach is that we can add NAS protocols (NFS and/or CIFS) and iSCSI in the future if we need to, all on the same storage system.
NetApp provides close integration with both Exchange and VMware, which can make the life of an administrator a lot easier. I recently participated in a NetApp TechTalk in which the specifics of many of these technologies was discussed. That gave me the opportunity not only to talk about what UBG was doing, but also to hear a lot more about NetApp and VMware solutions.
NetApp’s integrated Snapshot™ technology makes it easy to capture consistent, point-in-time images of Exchange data for backup, and it puts the backup workload on the storage system rather than on Exchange servers.
Snapshot is the foundation technology on which NetApp builds a variety of related services. For instance, NetApp SnapManager® for Exchange integrates with Exchange to coordinate actions between the Exchange server and storage to create backups that are consistent and fast. NetApp SnapManager for Virtual Infrastructure provides a similar capability to back up running VMware virtual machines. Because backups occur very quickly and put no load on ESX servers, you can create many more backups per day for greater data protection.
For disaster recovery, NetApp SnapMirror® technology builds on NetApp Snapshot to create a simple and reliable mechanism for asynchronous replication to a remote site. Unlike many other replication solutions, SnapMirror allows you to have different storage infrastructures at your primary and remote sites. You can use Fibre Channel HBAs at one and iSCSI at the other, and you can use faster but more expensive Fibre Channel disks at your primary site and less expensive SATA disks at your DR site.
Figure 4) NetApp SnapMirror provides flexible replication for virtualized environments.
NetApp also offers deduplication technology that works with your primary storage (rather than just on backups) to eliminate duplicate copies of blocks and reduce the amount of storage you need. This deduplication technology works particularly well in virtual environments because virtual machines are typically almost identical and thus have a large amount of duplication. The storage savings you get from deduplication of your primary site gets passed along to your DR site, reducing the storage needed there.
Figure 5) When NetApp deduplication is used at a primary site, the space savings are automatically passed on to the DR site.
NetApp has worked closely with VMware to integrate the features of SnapMirror with VMware Site Recovery Manager (SRM), a new tool from VMware that simplifies DR planning, implementation, and testing for virtual environments. (A companion article in this issue describes the integration of NetApp and VMware technology in SRM in more detail.)
The combination of VMware and NetApp technologies with Exchange 2007 has been a real win for us. Virtualizing our Exchange environment gives us greater reliability and flexibility with good performance. VMotion lets us move virtual machines easily for maintenance or other needs. DRS makes sure that Exchange servers are always properly resourced. VMware HA makes sure that Exchange and other important services get restarted right away should an entire ESX server fail.
NetApp storage is a good match for the capabilities of VMware, offering greater reliability at a lower cost than mirroring. We haven’t yet implemented all of the NetApp capabilities I’ve described above in our environment, but having a range of options will give us flexibility to meet our needs. We intend to look seriously at implementing VMware SRM using NetApp SnapMirror in late 2008 to provide greater protection for both our Exchange and Citrix environments.
Got Opinions About Exchange on VMware and NetApp?
While many IT teams are interested in running Exchange on VMware using NetApp, others are skeptical. What are your thoughts about combining e-mail with virtualized servers and storage?