Nice work. But be careful about relying entirely on something like this. There are more than just seek time differences between hard disks and SSDs: they have different internal cache behavior, different latency variances, queuing differences (depending on how the devices are configured and how you're using them) etc.
This seems like it'll get you 80% of what you want. But it'd be useful to have an actual disk to test on as well.
I think this is a great idea, but having worked on benchmarking database structures in the past, I'd be weary of using them for any type of real benchmark.
For one, trying to model a real disk would get very complicated very fast. Say, the access time of the disk will be a function of the position on disk, so it would be unrealistic to get random delays when scanning a large chunk of contiguous data or by just having a few delays in a random-access heavy load would also be extremely unfair.
In short, trying to model complicated disk latencies is pretty hard, and usually if you are programming with some model of disk in mind, building a disk latency simulator under that same model may end up giving you a false sense of security.
For what it's worth, I'd favor getting a cheap hard disk and trying the load there instead.
Comments
Nice work. But be careful about relying entirely on something like this. There are more than just seek time differences between hard disks and SSDs: they have different internal cache behavior, different latency variances, queuing differences (depending on how the devices are configured and how you're using them) etc.
This seems like it'll get you 80% of what you want. But it'd be useful to have an actual disk to test on as well.
To further this.
I think this is a great idea, but having worked on benchmarking database structures in the past, I'd be weary of using them for any type of real benchmark.
For one, trying to model a real disk would get very complicated very fast. Say, the access time of the disk will be a function of the position on disk, so it would be unrealistic to get random delays when scanning a large chunk of contiguous data or by just having a few delays in a random-access heavy load would also be extremely unfair.
In short, trying to model complicated disk latencies is pretty hard, and usually if you are programming with some model of disk in mind, building a disk latency simulator under that same model may end up giving you a false sense of security.
For what it's worth, I'd favor getting a cheap hard disk and trying the load there instead.