wait time , svc time, que depth, what is "interestingly" high in a generic sense

[Adam Garsha]

If you were drinking a beer at a baseball game, and a guy asked you:

What would you consider to be a high disk service time (in ms) or a figure that you'd be concerned that you were overloading the disk subsystem (a generic raid array), what would you say would be a number at which you would be "interested"?

Then what if he followed up and asked about an "interestingly high" level of i/o wait time (in ms)?

Finally, what if he said, huh, what disk system que depth would start to make you itch? (for mixed random/sequential i/o)

For me, I'd say:

I'd be interested at about 20ms for either of svc/wait times and really curious at 50+ms and when I was seeing 100ms, I'd think the system's i/o would be extremely sluggish. I'd say that I would expect to not see que depth over 5 most of the time and never above 20 for a long duration.

But I am rusty and I don't want to sound like a jagoff so I am seeking some grounding which I would not reference.

I know it "depends" (on a lot), but just looking for rules-of-thumb that drive you to look deeper.

[StuartA46]

Adam:

I agree with you.