justice@dao.nrc.ca (Gerald Justice) (02/09/90)
I am looking for some advice on optimizing the file systems on a Sun 4/280
NFS and boot server running SunOS 4.0.1 (soon 4.0.3) used in a scientific
data analysis environment. The clients are 3/60's and 4/60's, all diskless.
I have:
I) Read and re-read the mkfs(8), newfs(8), tunefs(8), and fs(5) man pages.
II) Read chapter 7 (the file system chapter) of _The_Design_and_Implemen-
tation_of_the_4.3_BSD_Operating_System_ by Leffler, McKusick, Karels,
and Quarterman. (Very enlightening but didn't answer my questions.)
III) Looked for but failed to find anything useful in the Sun "Network and
Administration Manual" even though all the above man pages refer there.
Specifically, I would very much appreciate:
I) A copy of _A_Fast_File_System_for_UNIX_, McKusick, Joy, Leffler. Can I
ftp it from somewhere? (This is also referenced in the Sun man pages.)
II) Advice on the effects of changing the following parameters:
A) number of cylinders per group
1) If one has a very large file system (order 700Mb) intended for
storing primarily large files (.25 to 2 Mb) although small files
will also occur in this fs, is there any point to increasing this
parameter?
2) If one has a partition for diskless client swap space (say 8 * 32 Mb)
is there any advantage to increasing this parameter?
3) If one has a very small partition, say 24 cylinders is there any
benefit in setting this parameter to 12 to have two equal
sized groups rather than two different sized (16 and 8) groups?
B) maximum blocks per group
1) as in A1) above, should this parameter be increased? How much
is safe? Note that efficiency of access for large files is more
important than for small files.
2) as in A2) above, is there any benefit to increasing this parameter?
C) minimum free space
1) as in A1) above, can less than 10% be "gotten" away with? 10% of
700Mb is a lot of "elbow" room, and seems inappropriate. But
read/write efficiency is the primary objective for this fs so I
don't want to sacrifice that to gain some more space.
(But I do so want more space!)
2) as in A2) above, is this really a "static" partition (like /usr)
and can I safely set free space to 0?
D) bytes per inode
1) I plan to double this parameter for those fs's where average
file size is fairly large and where similar partitions currently
show inode utilization below 30% but space utilization near 90%.
2) as in A2) above, is there any reason not to set the bytes per inode
to the actual swap file size? Should I really create just 8 inodes?
(Actually I would create 16 just in case, I suspect it is impossible
to get less than some minimum like blksize/inodes-per-blks?)
E) anything else that I can usefully adjust?
Thank you in advance.
Email responses will be summarized if there is sufficient interest.
justice@dao.nrc.ca BITNET:justice@nrcdao DAO Victoria, BC CANADA (604) 388-0055seeger@manatee.cis.ufl.edu (F. L. Charles Seeger III) (02/10/90)
In article <22401@adm.BRL.MIL> justice@dao.nrc.ca (Gerald Justice) writes: |I am looking for some advice on optimizing the file systems on a Sun 4/280 |NFS and boot server running SunOS 4.0.1 (soon 4.0.3) used in a scientific |data analysis environment. The clients are 3/60's and 4/60's, all diskless. | |II) Advice on the effects of changing the following parameters: | A) number of cylinders per group | B) maximum blocks per group | C) minimum free space | D) bytes per inode | E) anything else that I can usefully adjust? F) maxbpg (maximum number of blocks any single file can allocate out of a cylinder group---defaults to 25%) G) fragsize Specifically, how should these parameters be set for special file systems, such as ones that contain only exported swap files under SunOS 4, or news partitions, or others whose usage patterns vary greatly from the "norm". |Thank you in advance. |Email responses will be summarized if there is sufficient interest. I think that you may assume that there is sufficient interest if there is some good advice. A month or so ago I posted a similar request to a couple of mailing lists, but all that I received were "me-toos". I am interested in seeing either some anecdotal experience, real measured data, or the opinions of some real guru/wizard types. Thanks-in-advance. |justice@dao.nrc.ca BITNET:justice@nrcdao DAO Victoria, BC CANADA (604) 388-0055 Regards, Chuck -- Charles Seeger E301 CSE Building +1 904 392 1508 CIS Department University of Florida seeger@ufl.edu Gainesville, FL 32611
lm@snafu.Sun.COM (Larry McVoy) (02/10/90)
In article <22401@adm.BRL.MIL> justice@dao.nrc.ca (Gerald Justice) writes: >I am looking for some advice on optimizing the file systems on a Sun 4/280 >NFS and boot server running SunOS 4.0.1 (soon 4.0.3) used in a scientific >data analysis environment. The clients are 3/60's and 4/60's, all diskless. Hmm. I can't find the forest for the trees in this posting. Perhaps if you told me exactly what it is that you want to go faster I could help you. --- What I say is my opinion. I am not paid to speak for Sun, I'm paid to hack. Besides, I frequently read news when I'm drjhgunghc, err, um, drunk. Larry McVoy, Sun Microsystems (415) 336-7627 ...!sun!lm or lm@sun.com