iotop, I found out that [flush-8:0] is doing a lot of hard disk activity, which explains the annoying slowdowns. However, I can't figure out what's the actual cause behind this flushing. No other process seems to be doing inordinate amounts of IO. And you can't
strace a kernel task.
I've also enabled IO debugging (via
echo 1 >/proc/sys/vm/block_dump), but it gives me block numbers on a JFS filesystem, and I have no idea how to translate those into filenames (just try a web search to see what you get).
I actually have a suspect, a
java process that seems to cause hard disk activity whenever I interact with it. But
strace -eread,write,open,stat -fp $PID show very little activity. Are there any other syscalls that could cause problems?
Any other ideas?