Determining “equivalent power” / “equivalent processing power” / “equivalent computational ability” (for a processor)

Question

Certain software that we're planning to have on a dedicated server lists the following as the minimum requirements:

• Dual 2.6 GHz Quad Core Xeon i7 with hyper threading (16 total virtual processors) or equivalent power.

I think that the "Dual" means two here, which confuses me a little bit because I dont think a server will have TWO such processors (I may be mistaken though!)

(a) am I reading that requirement right? Does this mean that a single server has "two" '2.6 GHz Quad Core Xeon i7 with hyper threading' processors?

(b) how do i determine the equivalent computing power? There are numerous dedicated hosting providers with a wide range of processors. Is there any reference site which lists the comparative "computing" power of such processors so I may compare our requirement with what the provider provides?

Answer 1

1) A dual quad core (giving 8 cores in total) is a perfectly sane configuration for a server. Yes I think you are reading it right.

2) I don't think there is any way to accuratly judge equivelent processing power. It also depends on what the software has been optimised for. For example, your spec calls for a CPU that supports hyperthreading. If the software had been carefully optimised to make full use of hyperthreading then you would need to use double the number of non hyperthreaded CPUs to achive equivelent computing power. If on the other hand the software only has a tiny part that can make use of HT then you could probably miss it out, add a few 100Mhz to the clock speed and all would be well.
I would suggest contacting the supplier for help in judging what should be considered equivelent. If you are forced to guess then stick to modern (this generation or last generation) processors and compare based on number of cores and clock speed.

Answer 2

IME, recommended hardware requirements for a server are often swayed as much by where the vendor wants to position themselves in the market as they are by measurable requirements.

In effect that's 16 virtual processors working at 2.6GHz internally. But the maths is not as straightforward as that. Depending on the workload then the speed of the memory bus and availability of NUMA may be a determining factor. Also while software vendors often specify a minimum hardware requirement, in practice the capacity required for a specific level of performance(1) will vary by the amount of data(2) and its shape(3). Without significant research a vendor will not be able to determine the right values for 1,2 and 3 for each customer. Nor is it practical for the vendor to test every hardware combination against a pre-defined sample load.

A while back I decided to test out the Business Continuity support my employers were paying for at the time. The support company supplied a replacement system which was significantly faster (on paper) than the specification required (16X500 MHz processors compared with our requirement for 2x1000 MHz processors). I soon find out that we would be in a lot of trouble very quickly - part of the application concerned with batch processing ran in a single thread and took twice as long to run on the supplied machine. Problem was that it ran for 14 hours on our hardware and was required to run twice a day.

If you really want to compare apples and oranges then try costing an off the shelf server (an outright purchase of the hardware that is) proposed spec and the available spec - the market will tend to even things out - but do be aware that memory and disk are a significant part of the cost and should be as near as possible between the target systems.

C.