Microsoft has recently changed the basis of its server pricing from being CPU-based to being processor socket-based. This is good for Microsoft as well as Microsoft's customers.
Why has Microsoft made this change, and what does it mean?
Until quite recently, "Moore's law," which is a statement about the rate at which transistors increase over time on an economically producible silicon chip (they double every two years), has had a corollary -- the clock speed of a processor doubles every two years. Unfortunately, this corollary has now hit an electrical power, and therefore heat, wall. Observers will have noted that processor clock speeds have remained static for almost six years, and over three processor generations. But Moore's law has been marching on.
Since they haven't been able to increase clock speed, processor manufacturers have been increasing the number of physical and logical CPUs on a processor chip. The latest processors have eight to 16 logical CPUs per chip. For example, Intel's i7 has four physical CPUs, each of which is two-way hyper-threaded.
Another way of thinking about this change is that Moore's law has moved from operating in the temporal domain (clock speed) to the spatial domain (logical CPU count).
As long as Moore's law operated in the temporal domain, the number of transactions, users, etc. that could be handled by a server doubled every two years, while both the hardware and the software cost of that server remained static.
Once Moore's law started to operate in the spatial domain, the number of transactions, users, etc. that could be handled by a server continued to double every two years, while the hardware cost of that server remained static. Unfortunately, the software cost (if it was CPU based) also doubled. While at first sight this might appear to have been good news for Microsoft, in reality it wasn't. Microsoft's has always been a volume business. The company's mantra has always been: Lower the price by x% (where x is a large number) to increase volume by more than x%.
Microsoft's approach to this dilemma has been quite simple. Stop counting the CPUs and start counting the sockets into which processor chips plug. At a stroke, this returned Microsoft to a world in which the number of transactions, users, etc. that could be handled by a server doubled every two years, while both the hardware and the software cost remained static. A world in which the hardware and software costs of handling a transaction, user, etc. halve every two years, but the volume of transactions, users, etc. more than doubles. A world in which Microsoft's revenues continue to climb, while the unit-cost of a transaction, user, etc. continues to halve every two years.
Many other vendors are continuing with their CPU-based pricing. Expect them to adopt a similar, processor socket-based approach.
... Nick Shelness