Date: Thu, 23 May 1996 11:04:23 -0700 (MST) From: Terry Lambert <terry@lambert.org> To: scrappy@ki.net (Marc G. Fournier) Cc: sdd@ccd.tas.gov.au, freebsd-questions@freebsd.org Subject: Re: Missing Disk Space (df -k) Message-ID: <199605231804.LAA07821@phaeton.artisoft.com> In-Reply-To: <Pine.NEB.3.93.960523011512.21150B-100000@ki.net> from "Marc G. Fournier" at May 23, 96 01:16:50 am
next in thread | previous in thread | raw e-mail | index | archive | help
> > Okay.. Where has my missing 700Mbytes gone ? > > > > Anyone got any ideas.. Or is this just a symptom of a sick file store? > > > > Approximately 10% of a file system is reserved (set aside) for > fragmentation algorithms. That 10% *is* writable by root, but by > no other, so you can reach a negative value on it. It's 8%, with a recommended *minimum* of 5%, worst case. The reason for the reserve is because you are effectively hashing blocks onto the disk; according to Knuth (Sorting and Searching), the hash efficiency starts dropping off logarithmically at 85%. 10% was a good middle ground for a limit on degradation of hash efficiency. 8% makes people with *big* disks happy, of course, but doesn't change the mathematical principle involved. Any fill over a real 85% will start degrading performance, and any fill over 90% will probably degrade performance unacceptably. Actually, one could consider a change to the clustering algorithm to reduce the reserve by the average blocks in a cluster as a scaling factor. That is, if you didn't allow non-clustered blocks to be spread out all over (interfering with clustering) you could potentially divide the reserve by the cluster size. Terry Lambert terry@lambert.org --- Any opinions in this posting are my own and not those of my present or previous employers.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199605231804.LAA07821>