I imagine this benchmark is probably isolated from any disk read performance, in practice. The file sizes are small enough that the files are probably read into the page cache during the warm-up step, and then kept active, and stay there for the duration of the benchmark. Keeping the data decoded in main memory As long as the hosts aren’t showing a lot of page faults it’s probably not touching the disk often.
I believe so too. I did mention it. You know that feeling, when you've done any benchmark, and you step back and think about it and realize that your own benchmark is more and more questionable. :) If anything, it can be nice to know that it's not a massive difference worth worrying about.
Oh for sure—I think you mentioned not being sure how the page cache would impact it? But yeah absolutely. It feels so anticlimactic sometimes to come away from a benchmark and say: even with all the caveats, the result is: meh, the effect is small, decide another way
Comment
I imagine this benchmark is probably isolated from any disk read performance, in practice. The file sizes are small enough that the files are probably read into the page cache during the warm-up step, and then kept active, and stay there for the duration of the benchmark. Keeping the data decoded in main memory As long as the hosts aren’t showing a lot of page faults it’s probably not touching the disk often.
Replies
I believe so too.
I did mention it.
You know that feeling, when you've done any benchmark, and you step back and think about it and realize that your own benchmark is more and more questionable. :)
If anything, it can be nice to know that it's not a massive difference worth worrying about.
Oh for sure—I think you mentioned not being sure how the page cache would impact it?
But yeah absolutely. It feels so anticlimactic sometimes to come away from a benchmark and say: even with all the caveats, the result is: meh, the effect is small, decide another way