"proxy.process.cache.directory_collision": "57", We also see this counter going up when this event happens if that gives any clues.
Christopher Arnold [email protected] On May 2, 2012, at 8:49 AM, Bruce Lysik <[email protected]> wrote: > Ah, ok. So now it's even more different as we have gone back to 7 x 300GB > raw device drives instead of that one 1.7TB RAID device. > > -- > Bruce Z. Lysik <[email protected]> > From: Leif Hedstrom <[email protected]> > To: [email protected] > Cc: John Plevyak <[email protected]>; Bruce Lysik <[email protected]> > Sent: Wednesday, May 2, 2012 8:04 AM > Subject: Re: selectively clear device disk cache? > > On 5/2/12 9:00 AM, John Plevyak wrote: > > > > Means that you have too few directory entries for that disk. Odd, unless > > you are caching very small objects or have set the configuration for > > average object size too high. > Right, this might be my fault :). They have mostly large files, with 1.7TB of > disk, and the default of 8KB average size creates way too many directory > entries (wasting memory, disk I/O and longer startup times). They should > estimate how many objects they need, salt it appropriately, and tune the > config accordingly. > > -- Leif > > > >
