Performance metrics

Computer Performance Metrics

There are a number of ways of measuring the performance of a computer system or indeed that of the components that make up a computer system. One common measure is processor speed. The question of how to measure processor speed is not as simple as it appears.

 

One simple measure is the number of instructions that can be executed per second, expressed in millions of instructions per second or MIPS. So, a given processor might have a processor speed of 5 MIPS, i.e. it can execute 5 million instructions per second. But, all instructions do not take the same amount of execution time. An instruction to clear a register might take 1 clock cycle. A multiplication instruction might take more than 10 clock cycles. 

The way to compute the MIPS rate more usefully is to calculate the average time the processor takes to execute its instructions, weighted by the frequency with which each instruction is used. 

Another metric that is similar to the MIPS rate is the FLOPS or floating-point operations per second rate which is expressed in megaflops (MFLOPS) and gigaflops (GFLOPS). This metric is used particularly for machines targeted at the scientific/engineering community where a lot of applications software requires large amounts of floating-point arithmetic. Executed more quickly than others (e.g. addition versus division). Like the MIPS metric, the FLOPS metric needs to be approached with caution. 

The MIPS and MFLOPS metrics are concerned with processor speed. The processor is a crucial component of a computer system when it comes to performance measurement but it is not the only one and, in fact, it may not be the determining factor of the performance of a system. 

MFOPS 

MFLOP refers to Million Floating Point Operations Per Second, and is the traditional measurement used to describe the speed of a supercomputer. MFLOP measurement is based on normalizing the time of execution of a computer instruction to a constant amount of time when all instructions do not take the same time to execute. The variation in computer instruction time can vary between 29 and 600 computer cycles. MFLOP measure serves as a good approximation of the speed of a supercomputer as long as common baselines for instruction time of execution are well published if comparing speeds of two competing supercomputers. 

MIPS= n*1000000/t 

MFLOPS= 0.1 flop/inst * 1000 MIPS = 100 MFLOPS