home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.unix.ultrix:9076 comp.unix.misc:4712
- Path: sparky!uunet!spool.mu.edu!yale.edu!jvnc.net!princeton!rutgers!micro-heart-of-gold.mit.edu!news.media.mit.edu!news.media.mit.edu.!nessus
- From: nessus@media.mit.edu (Douglas Alan)
- Newsgroups: comp.unix.ultrix,comp.unix.misc
- Subject: Re: Ultrix newfs/mkfs -- can't change number of inodes?!
- Message-ID: <NESSUS.92Dec24135609@twitch.media.mit.edu>
- Date: 24 Dec 92 18:56:09 GMT
- References: <BzFv8K.JEJ@mtholyoke.edu> <1992Dec20.094432.9276@pony.Ingres.COM>
- Sender: news@news.media.mit.edu (USENET News System)
- Organization: Kate Bush and Butthole Surfers Fandom Center
- Lines: 35
- In-Reply-To: rog@Ingres.COM's message of 20 Dec 92 09:44:32 GMT
-
- > In article <1992Dec20.094432.9276@pony.Ingres.COM> rog@Ingres.COM
- > (Roger Taranto) writes:
-
- > If the block size is 8192, then the number of cylinders per group must
- > be a multiple of 16, and the frag size must be at least 1024.
-
- This is not really true. This limitation only occurs for some disktab
- entries. For other disktab entries, you can go down to a cylinder
- group size of 4 cylinders and maintain a block size of 8192. The
- conclusion from this is that the disk geometry parameters that you
- give to newfs effect the minimum cylinder group size. I've never
- determined the exact relationship here, but you can try giving newfs
- random disktab entries, until you find one that works.
-
- "Doesn't giving newfs a random disktab entry mess up the fine tuning
- parameters for the filesystem and thus degrade performance?" you might
- ask. Well, perhaps it does, but I've done a fair number of empirical
- tests and never noticed much of a differenence. Something that has a
- much bigger effect it usually to up "maxcontig" using tunefs. This is
- not done by default and can cause a significant performance increase.
- (it also might have no effect, or decrease performance if set too
- high, so you should do some measurements.)
-
- You may think that setting the block size down to 4096 is a better
- alternative than messing up the fine tuning parameters. This is
- probably not true. The two most significant reasons for the good
- performance of the Berkeley Fast File System is the large block size
- and the fact that seeking distances are reduced by keeping most
- adjacent blocks in a file in the same cylincer group. Reducing the
- block size to 4096 would be diminishing one of the most important
- sources of the filesystems performance gains. The gains from the
- fine-tuning parameters are relatively minor in comparison.
-
- |>oug
- <nessus@mit.edu>
-