An alternative to bogomips Running 10 iterations of computing the cryptographic hash of 512 copies of the system default dictionary, we can measure the relative performance of a variety of machines snd compare that to the use of Bogomips as a proxy for system performance. Experience with this test that combines computationally intensive tasks that have modest amounts of reading and including a "dd" to /dev/null of the same input files, we get a rating (by computing geometric means) that have matched this author's intuitive sense of the relative performance of the machines.
The process of testing a computer system is much more complex than just getting a measurement of CPU perormance. This set of scripts were motivated by the fact that the Linux Foundation in their "ready-for" scripts were using bogomips as a proxy for system performance.
- Bogomips was never intended to be used this way.
- Bogomips does NOT measure system performance, only the speed of a small timing loop.
- Bogomips is broken with respect to CPUs that use variable clock rates
This is an attempt to create a proxy for system performance. In the process of creating this I explored some alternative ways of performing the metric. The default dictionary is ~ 1 MB (976,241 bytes) in size. When you run 10,000 to 15,000 iterations over the dictionary using either a cryptographic hash program (bssum, sha1sum, sha256sum, sha512sum) or the dd application it becomes easy to see that the time spent loading the applications into memory will quickly dominate and have a skewing effect on the results.
By making 512 copies of the dictionary, the size is now 499,835,392 or just under 1/2 GB. Using this larger file improved the performance of all of the applications.
To run a set of tests on A Machine, run the script: dohosts -o and find the results in the results directoy.