Sustained io testing linux

Question

I wish to hammer the io of some disks for an extended period of time and become aware whenever I can't read from a block (or some other symptom letting me know of an issue with backend storage). There are some benchmarking tools that write for a few seconds and show you results, but I want to do long-term testing.

So far what I can think of is writing to disk via dd and reading from that file to /dev/zero . I would need to loop it so as to keep reading and writing after it has finished the initial run. As for having an insight into disk health, I suppose dd may terminate if it can't read or write? Otherwise I may not know if there is an issue.

The other idea is to run bonnie++ in a loop. It's hard to tell what's doing in the background and how much ram it is actually using instead of disk (seems they try to get around this by telling you to have a large data amount written; larger than your ram allocation). Then, the output it gives you is pretty hard to read. But this should suffice for writing and reading if I use a bash loop to run it constantly.

Thoughts?

`I suppose dd may terminate if it can't read or write?` Not **may**, but will. — Michal Sokolowski, Jun 18 '15 at 01:18
What do you expect to find? What is the purpose of this testing? — ewwhite, Jun 18 '15 at 03:04
@ewwhite verify the integrity (reliability) and sustained performance of the backing storage. — derpy, Jun 18 '15 at 03:51
For it to work properly you would have to disable read cache. Both on VFS, and on the drive itself. — Konrad Gajewski, Jun 18 '15 at 09:32

score 1 · Answer 1 · edited Apr 13 '17 at 12:14

How about an fio job with verification and a time_based workload? I'm thinking something like

fio --name=2h --direct=1 --filename=/dev/sdz --verify=crc32c-intel --verify_fatal=1 --time_based=1 --runtime=2h --rw=write

This will do a random write of /dev/sdz bypassing the block cache (thus destroying the contents of /dev/sdz). After the /dev/sdz has been fully written its contents will be read back and verified. This process will be continued for two hours and the previous steps will be repeated each time they finish. This is obviously imperfect (e.g. there's no guarantee even the first write pass will have been completed by the end of the two hours etc) but it gives some ideas...

If you're doing some sort of storage I/O testing then fio is often worth look.

Sustained io testing linux

1 Answers1