In a recent post I described the last "pre-virtualized" deployment of a Microsoft Exchange configuration for EMC Corporation. IT experts had designed a system that could scale to an impressive 48,000+ mailboxes. Clearly, that configuration does not satisfy the cloud characteristics that would eventually be defined by NIST. In addition, the single-tenant Exchange deployment sat alongside (and separate from) other single-tenant deployments (e.g. Oracle). Dedicated servers, networks, and storage began to sprawl and consume increasing amounts of electrical power.
As part of the IT Production phase, EMC IT aggressively pursued server virtualization. This improved IT efficiency on several fronts. The pace of lab growth slowed, less power was consumed, and decreased spending on tech and server management was realized. Non-business critical assets owned by IT that were viewed as "low-hanging fruit" were virtualized first. In addition to server virtualization, IT was systematically improving efficiency in other areas, consolidating islands of infrastructure, converting to disk-based backup, etc.
What about the virtualization of mission critical apps like Exchange?
The Business Production Phase
A virtualized Exchange architecture would improve the quality of email service received by EMC employees. Working with VMs would be faster and more efficient. The physical clustering of dedicated Exchange servers would be replaced by the logical clustering of Exchange VMs. EMC Distinguished Engineer Wissam Halabi participated in the construction of the new virtualized architecture. The pictures below contrast the old and the new.
The number of servers in the top configuration is dramatically greater than in the bottom. The key to this efficiency is magnified by zooming into a single vSphere Exchange node and comparing it against the physical mailbox server in the old configuration:
The virtualized solution hosts 16,000 mailboxes on one server, compared to 12,000 mailboxes on three. It's a clear IT efficiency win. Note also that the two-tiered storage configuration (DMX and CLARiiON) were consolidated down to a replicated pair of VMAX systems by performing application migration via Exchange's mailbox migration wizard.
Certainly this configuration does not satisfy the NIST cloud characteristics, but it is a step towards a more fully virtualized data center (VDC). Indeed, at this point in EMC's deployment the percent virtualization was climbing, and the quality of service was improving to the point that the company as a whole had entered into the business production phase:
As the graphic above indicates, the virtualization of mission-critical apps (like Exchange) improves IT's overall quality of service. For example:
- Deploying a new mailbox cluster is easier and faster with the ESX solution.
- Allocating new storage and assigning it to a new ESX node is much less complex (only one tier of storage: VMAX), and the provisioning/assignment of LUNs to virtualized server ports can be streamlined via the use of templates.
- Replicating this configuration is also simpler (only using SRDF as opposed to both SRDF and CLARiiON MirrorView)
- Blackberry VM's were introduced into this configuration, providing broader network access on more devices.
The OPEX of the solution is vastly improved due to the more homogenized configuration. VMAX also can provide other benefits that are typical of the business production phase: integrated management and security, as well as the ability to provide tiered storage via FAST.
What are the next steps being taken to move this configuration towards true business agility? I'll cover that in future posts.