>>> "Andres Kroonmaa" wrote
> FD usage is quite similar for both versions with only NOVM/MISS
> having 50% more FD's in use, while memory use differs by several times.
And this is the critical difference for us. Our caches are creeping
ever-closer to the 4096 FD limit of Solaris as their load increases -
we can always just add more memory to the box (well, we can add enough
that it's not going to be a problem), but extending the number of FDs
doesn't seem to be possible (2.6 doesnt seem to help :-(
> One thing is _average_ which is nice and useful thing for decisions,
> but the other thing is _peak_ usage, and thats what NOVM deals better with.
It depends on what your limiting factor is. If it's memory, the system goes
slower. If it's filedescriptors, it just stops.
If people are downloading big files, surely inactive parts of the big files
will be paged out, and left out in swap until after the file is fully
retrieved? Ah, mallocs with metadata in with everything else are going to
still have to page the stuff back in... hmm, has anyone looked at using
Doug Lea's malloc with squid? <URL:http://g.oswego.edu/dl/html/malloc.html>
Received on Tue Jul 29 2003 - 13:15:41 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:18 MST