Virtual machines: the new ghetto in the data center

Virtualization is becoming a ghetto for some lines of business applications. Those server-based applications, each of which used to get its own dedicated hardware, are increasingly stacked like pancakes on a single machine. In some companies, applications that used to have their own homes are now housed in the IT equivalent of cheap tenements. In some cases 50 or more applications may reside on a single physical server. But with too many virtual machine tenants-in-residence, the underlying hardware foundation starts to crack. Performance degrades. The user experience begins to suffer.

It wasn't supposed to be that way. Management, troubled by server utilitization rates in the range of 10%, was delighted to discover virtualization as a way to get more out of existing hardware. Some organizations have begun cost saving initiatives that require all new applications to reside in virtual machines unless a business case can be made to break policy. Unfortunately, the temptation for some overzealous bean counters is to put too many applications on each physical server.

It used to be that every application would get its own physical server, and since hardware is cheap and powerful, users were ensured of snappy performance. Users had more power than they needed. Unfortunately, some organizations may be stepping back into a world more familiar to mainframe users in the '60s and '70s. In that era of timesharing, applications got a slice of the mainframe. Users often had to wait for their jobs to complete because a high volume of jobs was competing for scarce and very expensive hardware processing resources. Timesharing was a compromise, a way to allocate those precious MIPS.

Wintel servers made processing power cheap and plentiful, but the limitations of one application per physical server helped promote server sprawl.

Now for the first time IT can consolidate those compute resources and, at a very granular level, allocate them to more closely match each application's needs. While the cost of processing power has dropped, the move to virtualization has allowed management to dramatically throttle back the allocation of that processing power to each application in the name of cost savings. The irony is that, in a world where processing power is cheaper than ever, some users may find themselves waiting longer.

So once gain, users are competing for compute resources. How big a slice does your application get? And how slow is too slow?

As a consumer of compute services, that may depend on how important your application is to the bottom line and how much political clout you have in the organization. For one user, who told me he had to wait for more than a minute for even a simple SQL query to complete, the practice would seem to have gone too far. The usage pattern of the application - very high CPU utilization rates - would seem to dictate the use of standalone hardware. But getting on a dedicated server practically takes an act of God because it breaks with the policy to use virtual machines. In this case, the burden is on the user to prove that the performance issue was real and that it was a productivity problem.

Analysis and planning tools are available to properly populate server hardware with virtual machines and allocate resources to them. Dynamic load balancing tools are rapidly evolving as well. However, organizations are still climbing the learning curve as they adopt virtualization technology. In the short term, there are likely to be abuses. In some companies the pendulum will swing toward penny-wise and pound foolish as bean counters demand tighter budgets. IT will struggle to figure out just how much of the computing pie a given applications deserves - and what performance levels are good enough.

Such determinations can't always be made purely on a technical basis.

The core technology of virtualization has matured. It may take a while for best practices to catch up.