Quick edit to my post on virtualization - Information Week published an overview of the market and key drivers recently. It is worth reading.
Utility computing is a hot topic. Companies are spending millions of dollars on VMWare and other virtualization technologies in order to consolidate data center operations and pool IT resource demand. Utility computing and virtualization are reshaping today's data center technologies and operations. Why?
Quite simply, virtualization technologies drive significant improvements in asset utilization by pooling resources across variable demand. By pooling fluctuations in demand, one needs fewer physical resources to meet resource demand. Fewer provisioned resources reduces operating and capital expenses, a good thing:)
The technology industry is borrowing an important operational best-practice from supply chain practioners - that is the pooling of inventory. Why does this work?
It works because averages are additive, while standard deviations are not. This statistical fact drives the benefits of pooling inventory (IT assets), across variable demand for those assets.
In operations, in order to avoid lost sales, supply chain managers pad inventory levels in order to avoid expensive stock-outs. The IT equivalent is to overprovision. Supply chain experts leveraged basic statistics to realize that by pooling inventory one can greatly reduce the amount of safety stock required to adequately meet demand.
How? A simple (VERY) example shows the power of pooling.
Assume two applications, managed independently, with the following weekly demand characteristics:
Mean Demand 50 servers
Standard Deviation 4 servers
Mean Demand 50
Standard Deviation 3 servers
If we set safety stock equal to two times the standard deviation, application I needs 58 servers to adequately meet demand (ie with 95% confidence), while application II requires 56 servers. The data center responsible for both applications, in aggregate, overprovisions by 14 servers of "safety stock."
Now if we decide to pool the two applications, we would observe the following:
Mean Demand 100 servers (averages are additive)
Standard Deviation 5 servers, or sqrt(4^2 + 3^2)
Pooling the server demand of the two applications allows the enterprise to reduce server safety stock by 4 units, or by 28%.
For a large enterprise, reduced server safety stock translates into large savings.
Theoretically, 4 servers x 1,000 applications x $5,000 per server (estimate) = $20,000,000
This is an artificial example, but I think it is illustrative of the value of virtualization. It helps me, at least!, tie supply chain management operations theory to high-tech market adoption.
By leveraging enabling technologies to pool spare capacity and statistics (averages are additive but that standard deviations grow by the square root of the sum of the standard deviations in question), IT operators can continue to strip capacity and costs out of the data center while continuing to meet demand.