Date: Mon, 04 Mar 1996 17:01:03 -0800 From: David Greenman <davidg@Root.COM> To: Terry Lambert <terry@lambert.org> Cc: msmith@atrad.adelaide.edu.au (Michael Smith), monboso@masternet.it, freebsd-hackers@FreeBSD.ORG Subject: Re: xfs not working properly Message-ID: <199603050101.RAA19104@Root.COM> In-Reply-To: Your message of "Mon, 04 Mar 1996 17:19:19 MST." <199603050019.RAA07035@phaeton.artisoft.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>> > The DOS clients are generally broken, unless they have local cache, >> > but the FreeBSD NFS server is also "much slower". >> >> Hmm. I'll agree that NFS servers are, in general, not so snappy, but the >> FreeBSD server does a pretty good job, really. It's certainly not responsible >> for the atrocious performance the original poster was complaining about. > >Wasn't my point. My point is that there is room for improvement. > >I'd like to see variant LRU insertion order as well, FWIW, based on >a per vnode working set quota to prevent mmap'ing and sequentially >accessing an mmap'ed file (ie: "cp") from thrashing the cache. When >you hit the quota, you steal from your vnode instead of from the >global LRU (need to use a bit, like VEXEC, to defeat this behaviour >for shared pages, though...). John and I have discussed this before. Basically, use vadvise(sequential) to inform the kernel that operations are primarily sequential, and then use some sort of per-vnode working set quota to throttle the memory consumption. ...but right now, we've got lot's of serious bugs to fix so projects like this only get worked on when we're looking for a break (e.g. low priority). -DG David Greenman Core-team/Principal Architect, The FreeBSD Project
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199603050101.RAA19104>