How VMware Virtualizes Its Exchange Server Environment
VMware itself uses vMotion and vSphere to ensure availability and optimal performance from its Microsoft messaging system.
Ask Alex Fontana about running Microsoft Exchange on the VMware virtualization platform, and he all but calls it a marriage made in heaven. Microsoft has expanded its support for virtualizing its mission-critical communications backbone with VMware Inc. since it first offered a nod to the idea with the release of Exchange Server 2007 SP2. With the release of Exchange Server 2010 SP1, Redmond expanded its support for features such as the vMotion live migration tool and the vSphere high availability (HA) solution to be used along with some Exchange features. But like any marriage, this one takes a little work to succeed.
Exchange architects need to understand the design considerations and options available to them when choosing VMware as their virtualization platform, Fontana says. The senior architect in the VMware Solutions and Service group knows what he's talking about: He runs the team at VMware charged with virtualizing the company's own Exchange environment for about 12,000 users.
"When we get involved in virtualizing an Exchange environment, we generally opt for the least amount of hardware and the least amount of VMs -- the keep-it-simple approach," Fontana told attendees at the recent VMworld 2012 conference in San Francisco in a session called "Virtualizing Exchange Best Practices." But, he added, "there are a lot of ways to skin this cat."
Fontana also outlined common methods for provisioning storage in the VMware vSphere virtualization platform. VMware Virtual Machine File System (VMFS) is a cluster file system used by the company's flagship ESX Server. Its VMFS-based virtual disk is a high- performance, clustered file system designed to allow concurrent access by multiple hosts to files on a shared volume. The company's Raw Device Mapping (RDM) is a mapping file within a VMFS volume that acts as a proxy for a raw physical device. (RDMs are also known as "pass-through disks.") The RDM file contains metadata used to manage and redirect disk access to the physical device, he said, and RDMs can be Fibre Channel- or iSCSI-attached.
VMware defines a "thick" virtual disk as one that occupies an entire fixed amount of provisioned storage space. A "thin" virtual disk takes up only as much of the provisioned storage space as needed for its initial operations, growing into the rest of the space later.
Microsoft supports all of these technologies in a VMware virtualized Exchange environment, Fontana noted. It supports virtualization of all roles, including the Lync Server, and combining Exchange Server 2010 database availability groups and hypervisor HA (vSphere HA).
But there are a few things Microsoft doesn't support in this context, including NAS, VMware thin virtual disks and the use of VM snapshots for backups.
Fontana is frequently asked about VMware Data Recovery (VDR), or other tools that use a snapshot to back up the VM. His response? "VDR wouldn't be supported because it's not an Exchange-aware backup," he said. "But there are other vendor tools [supported tools include products from Veeam, and Quest Software vRanger] that take a VMware snapshot and employ agents within the guest to quiesce Exchange, truncate logs and do the verifications that are required when a VSS [Windows Volume Shadow Copy Service] backup is taken. Those are supported, because they're Exchange-aware."
Fontana shared a number of VMware-recommended best practices for virtualizing Exchange with vSphere. Topping his list: No over-commitment -- at least during the design phase and the project pilot, and even sometimes into production. Why?
"Although the Exchange storage calculator has a little dropdown that asks whether you're going to virtualize or not, the calculation doesn't take into account any level of consolidation of the physical cores themselves," he said. "It's basing those numbers on dedicated cores -- on the [expectation] that Exchange will be the only workload living on that core."
It's actually adding a fudge factor or buffer of about 10 percent for what's called "hypervisor overhead," he added. "At VMware, in our tests of Exchange 2007 and vSphere 4, we saw between 2 percent and 5 percent of that overhead," he said. "If you want to build that overhead into your design, that's OK, but we typically don't, and we see even less overhead now with vSphere." Hence, when running these calculations, "we want to make sure that we're not factoring in any over-commitment," he explained.
Another best practice: Enable hyperthreading. This is a tough one for Exchange admins, Fontana said, because Microsoft advises against it. Redmond argues hyperthreading makes designing and sizing extra challenging, because it confuses people about the actual compute resources that are available.
"We don't think this is the case -- or it doesn't have to be," Fontana said. "As long as you understand that you're not deploying against 24 physical cores, that you really have the throughput of only 12 physical cores, you get a significant advantage."
The advantage: Having those logical processors available will allow ESX (the bare hypervisor) to make more-intelligent scheduling choices.
Also a highly recommended best practice: Enable non-uniform memory access (NUMA) and keep the VM to within the NUMA node size. Enabling NUMA and keeping the VM small can provide a performance increase of about 5 percent, Fontana said. "If I can create a VM that fits within a NUMA node, ESX will do its best to keep that VM scheduled on those processor cores, thereby keeping its memory access local."
On the question of whether to use the VMware Paravirtualized SCSI (PVSCSI) or LSI Logic virtual adapters, Fontana came down firmly on the side of the latter. "It's the default and it always passes [our tests] with flying colors," he said. "We've seen scenarios where customers want to consolidate a large number of mailboxes on single VMs and they really need to drive some high I/O. But when sized properly, the LSI Logic adapter can do the work. I say base your decision on what you're running today."
Fontana pointed to a Web page where VMware has published material about virtualizing Exchange with its platform. The page includes links to blogs and a community portal.
John K. Waters is the editor in chief of a number of Converge360.com sites, with a focus on high-end development, AI and future tech. He's been writing about cutting-edge technologies and culture of Silicon Valley for more than two decades, and he's written more than a dozen books. He also co-scripted the documentary film Silicon Valley: A 100 Year Renaissance, which aired on PBS. He can be reached at firstname.lastname@example.org.