Do you any idea, how do I use IOZone on a customized file system, where there are specific APIs exists. Like xxx_open, xxx_close, xxx_read, xxx_write …

Is porting iozone is the only solution to support new FS? Does anybody ported already?

regards
Prasad

Please read the first article – it explains why I chose the record lengths.

Jeff

Ahh… grasshopper, welcome to the fray. The question you ask is the perfect one – how are benchmarks relevant to my workload. The answer is that – it ain\’t easy and usually impossible to determine that.

One of the projects I\’ve been working on is a way to do IO profiling of your application (look at some of the early columns I wrote – it\’s called strace_analyzer). From the output of this tool you can get an _idea_ of whether your app is IOPS driven or perhaps throughput driven. It also tells you the size of the records during reads and writes.

So \”theoretically\” with this information you can look at benchmarks that sort of mimic that behavior.

What I\’m working on next is to take the strace information and actually create a pseudo-application that just mimics the IO behavior. Then you have a real benchmark – your application. (I can\’t solve the problem of test hardware though – I\’m trying to solve that problem for me personally :) ).

Jeff

Thank you for this series of articles, they\’ve been very enlightening.

One suggestion to possibly avoid memory cache effects and surely reduce the time needed for the tests could be using the \”mem=\” kernel parameter so you don\’t have to pull the DIMMs out.

That way you might be able to use a smaller file and smaller record sizes but still get relevant numbers from the benchmark.

Thanks for the suggestion. It definitely would help things :) These benchmarks take a great deal of time to run :)

Thanks!

Jeff