Date: Tue, 14 Aug 2007 22:23:56 -0500 From: Eric Anderson <anderson@freebsd.org> To: Gore Jarold <gore_jarold@yahoo.com> Cc: freebsd-fs@freebsd.org Subject: Re: help needed - tuning a filesystem for rm and cp ? (more details) Message-ID: <46C271CC.8080008@freebsd.org> In-Reply-To: <924334.71021.qm@web63006.mail.re1.yahoo.com> References: <924334.71021.qm@web63006.mail.re1.yahoo.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On 08/14/07 16:47, Gore Jarold wrote: >>> Here is a question for any and all out there > reading >>> ... what would you expect would happen to a system >>> that was constantly maxing out this value, > sometimes >>> on a sustained basis, while the activity that > caused >>> it went on uninterrupted ? >>> >>> I am seeing the system halt ... is it reasonable > to >>> think that maxing that value out on a regular, >>> sustained basis would cause a system to halt ? >>> >>> (6.2-release running on a 4 GB memory p4 xeon ... > does >>> nothing but fileserver duties) >> >> If you have a lot of meta-data IO (which you seem to > have), then it's >> possible that the system is incredibly busy doing > disk accesses, and >> waiting on IO from storage. When you say 'halt' > does that mean you >> can't log in to it, and eventually it comes back > alive, or does that >> mean it is locked up in a way that never recovers? > > > I never wait around to find out it it comes back ... > > It no longer responds to pings, so I assume it is > actually crashed. > > Also, it should be noted that this system (and its > large, >4TB filesystem) has quotas enabled. > > So perhaps a better question would be: > > Any comments on frequent, sustained maxing out of > dirhash on a quota'd filesystem ? If you can, it would be best if you can break into the debugger and get a core. I can't think of how dirhash would effect quotas really, but I suppose it might be possible. You might try bumping the dirhash setting way up, maybe to 20MB or so. I think it might be worthwhile also to run a script (cron maybe) every minute that logs the current use of dirhash, and maybe a ps -auxwl, with a date stamp so you can when your peaks occur, what was running, etc, just before a crash. Might also look into bsdsar - it might help with some of this too. Eric
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?46C271CC.8080008>