I'm running a variant of Debian (Proxmox Virtual Environment) on my HP server, and I've noticed a strange quirk: while performing a backup to a USB drive (usually around 20 gigabytes worth of data, so it takes a while), naturally I experience a lot of IO delay. As seen by top (listed as "%wa"), as well as Proxmox's built-in monitoring tools (listed as "IO delay"), this delay can be as much as 50% (whatever that means)

So physically, my bottleneck is that particular USB hub that the drive is connect to. But the server itself starts to become unresponsive at this point. SSH sessions lock up, read/write commands to the internal drive freeze (the internal drive is definitely not the bottleneck), and network activity drops to a minimum.

My question is this: why, if only a particular part of the computer (the USB hub) is saturated, is the whole system held up, even the parts that are completely unrelated to the hub? Am I misinterpreting exactly what IO delay is? I assume that it's just the read/write delay to that particular drive, but I must be mistaken.

  • @Ignacio Agreed, but not exactly what I was looking for lol. USB is my first choice only because of how often I need to disconnect and move the drive – TheFuzzyFish Apr 30 '18 at 4:07
  • 2
    May or may not apply to you: I've been experiencing similar slow-downs under heavy I/O, dropping caches (echo 3 | sudo tee /proc/sys/vm/drop_caches) solves the problem at least temporarily. I suspect deadlocks when multiple cores access the I/O buffers, but haven't found anything concrete. – dirkt Apr 30 '18 at 5:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.