« Virtualization of Enterprise Systems Virtualized Servers | Main | Virtualization of Enterprise Systems - Index »



Feed You can follow this conversation by subscribing to the comment feed for this post.

Douq Millar

Virtualization (cloud) is often compared to the electrical power grid, but there are important differences.
With dedicated resources, providers knows their costs, customers know their SLA – both are essentially fixed over short-term.
The whole point of virtualized hosting is load sharing, not every customer is using their maximum resources at the same time, so provider can build out to some “average” which presumably costs less than dedicated. The provider spends less money for same # of customers, so can charge less (sales advantage) and get higher margins: win-win, who wouldn’t want that?

Now given virtualization does carry some premium in cost for the same delivered resource + provider has to have some headroom for customers peaking their loads at same time, provider’s savings may not be as much as purely the load averaging, esp. if the provider must define some SLA to customers. So operators cost savings may not add up to that significant a difference as it first seems and customers may see degraded service.

But here’s where the analogy to electrical grid breaks down. Electricity production economics is based on both capex and opex (fuel, labor, maintenance, etc) where opex is very high. In the summer afternoons in California when business still has lights and AC going full blast, but workers head home and turn on their lights, appliances, AC – loads peak. What do utilities do? Brownout, not very popular. Standby capacity that can be switched on quickly, better. Standby has same capex as baseline (actually usually more), but much lower (average) opex so meeting peak demand is not so hard for electrical grids (as long as Enron is not involved).

But any concept of “standby” in computing is vague since costs are almost entirely capex. What savings are there to “turn off” some vBlock? Not much. And, AFAIK, grid load sharing across service providers doesn’t exist so a provider can’t do want California does, buy electricity from Nebraska.

So inevitably there is a collision between SLA and costs. In some time periods (month-end?) load will rise for all customers. And since computing often has the nasty characteristics that pushing beyond some optimal load actually decreases performance (due to congestion itself imposing a load) Brownout time. So as a service provider where do I set my peak capacity? And then what are my cost savings to offer better prices and get me better margins?

So I think the airlines are a better model. They overbook, hoping not everyone shows up. Then they either bribe some customers not to fly or they delay everyone (brownout). Airlines don’t usually have spare planes and crews they can turn on at peak. So does service provider offer some second-class service where customer gets bumped when loads get too high, maybe. As customer can I live with that to get a little cost savings.

So the power utilities aren’t perfect, but frankly they do better load management than the airlines. Which model will virtualized server providing end up following?

Douq Millar

I want to trademark “Peak++” because IMHO this is the win, not cost-savings, of virtualization for service providers.

Instead of overbooking (airlines) or balanced-booking (most business) of resources, providers deliberately underbook, thus having ample reserve capacity to offer Peak++, something that is difficult to do with dedicated hosting. Economy customers get brownouts, first-class gets all-they-can-eat.

Roughly the same average cost as dedicated but now a premium service offering as the provider’s competitive advantage.

Today most providers are in this mode anyway, since they’ve built out anticipating demand and haven’t yet hit steady-state (and probably lose money betting on future growth). But five, ten years from now this startup headroom will be gone and steady-state will apply. So what does virtual service look like a decade from now?

managed it support services

This post shows the distinction between some of the popular hosting topologies and elucidate their practical application while drawing a comparison....I appreciated the way have distinguished the hosting services.

The comments to this entry are closed.

Cloud Computing Journal
Top 100 Cloud Computing Bloggers