Without proper controls in place, some IT professionals that deploy virtualization can quickly become victims of...
their own success, struggling to keep pace with demand for new virtual machines (VMs).
Managed service provider and IT outsourcer Computacenter U.K. Ltd.. started out with virtualization by doing physical-to-virtual (P2V) migrations of some 50 internal systems, with plans to P2V another 100. A year later, its virtual infrastructure runs at capacity with 250 VMs, but the IT team never got around to finishing the P2V migration of the old systems.
"There was a real focus on deploying new systems," said Martin Clarkson, a virtualization technology champion at the firm, who said that demand for new VMs came from internal developers as well as from new managed services customers.
Publishing giant the McGraw-Hill Cos. also witnessed explosive growth of its virtual infrastructure, but for different reasons. In 2008 the firm began offering internal business clients the option of a virtual machine "service offering," a single "slice" of compute, memory, network and storage capacity that users could rent by the month. The firm started out with three VMware ESX Server hosts in 2008 running about 20 VMs. A year later, it runs 75 hosts and 800 VMs, and the environment is on track to double in the coming year.VM sprawl tests IT management
In terms of adoption, both firms' virtualization deployments were a smashing success. But for the IT staffers manning the infrastructure, things quickly spiraled out of control.
"The actual number of people managing the infrastructure is quite small," said Computacenter's Clarkson. "They very rapidly got in a position where they couldn't cope with all the requests coming in."
McGraw-Hill was also caught unawares. "We weren't expecting it to go quite as quickly as it did," said David Hansford, the lead engineer for global operations and infrastructure.
Faced with this sort of unbridled growth, both firms moved to restore some sense of order.
At Computacenter, the solution was to deploy Embotics V-Commander, a virtual inventory management, control and automation suite.
"As I see it, [V-Commander] gives IT two things: a level of control over the VMs and whether they are valid and should be there; and an audit trail of where the VM came from," Clarkson said, adding that a lot of VMs are cloned by developers from existing virtual machines, and frequently do not need to be kept.
Embotics recently released V-Commander 3.0, which divvies up its core capabilities into separate, related modules, and adds Active Directory integration.
At McGraw-Hill, VM sprawl is reined in somewhat by placing strict limits on the scope of the virtual service offering. For instance, the operations team offers only VMs running Windows Server 2003 and Red Hat Enterprise Linux (RHEL) 4 and 5. Nor will it perform P2V migration. "We don't want any legacy NT 4 servers, because then we'll never get rid of them," Hansford said.
Even with these limits, the service offering is still more compelling than the old model, in which it would take up to 90 days for individual business units to procure a new server, at a cost of $10,000 to $30,000, and require a three-year commitment. In contrast, McGraw-Hill will deliver a VM with 1 GB of RAM, 28 GB of disk space, 1Gb/sec network throughput and unlimited CPU usage for a flat fee of $200/month, with no more than two weeks turnaround time.
The team also put in Vizioncore's vFoglight for monitoring and vControl for self-service provisioning and automation. "We had a requirement for some sort of monitoring," Hansford said, as well as billing, chargeback and provisioning. "We wanted a product that could do three out of four of these things."
McGraw-Hill uses vControl as the interface that internal customers use to submit requests for a VM, which are then parsed off to the IT team and propagated into VMware vCenter and the chargeback system.Tiering VM services
Meanwhile, demand for virtualized environments continues, pushing Hansford to explore ways to expand VM service offerings. One idea is to offer second-tier service that would not run on an N+1 cluster, but that would allow P2V migrations, possibly for a limited lifespan and for a higher fee. As it stands, McGraw-Hill's standard N+1 cluster running VMware and Vizioncore costs IT "six figures before we even start to get any money back on it."
Hansford is also considering extending its use of vControl to allow VM owners access to their VM consoles, without having to give them access through VMware vCenter. "It's not part of the service offering now, but obviously people have a need to be able to stop and start their machines."
Let us know what you think about the story; email: Alex Barrett, News Director.