3 ms·
Not in this case, it's all mmap.
by marginalia_nu 4y ago
Not in this case, it's all mmap.
- winrid 4y agoReally? Even with mmapp'ed memory won't the CPU still register user code waiting on reading pages from disk as iowait? I'm so surprised by that that if it doesn't it sounds like a bug.
- marginalia_nu 4y agoYeah it's at least what I've been seeing. Although it could alternatively be that a lot of the I/O activity is predictive reads, and the threads don't actually stall on page faults all I/O that often.