home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!think.com!spool.mu.edu!agate!ucbvax!lrw.com!leichter
- From: leichter@lrw.com (Jerry Leichter)
- Newsgroups: comp.os.vms
- Subject: re: REPOST: Parameters for Diskperformance
- Message-ID: <9301111414.AA08411@uu3.psi.com>
- Date: 11 Jan 93 12:52:08 GMT
- Sender: usenet@ucbvax.BERKELEY.EDU
- Organization: The Internet
- Lines: 85
-
-
- ..... I think the first time I selected distribution:world, so here it
- comes again......
-
- Well, I don't recall seeing this before, and I'm in the world somewhere....
-
- You also carefully included a complete mailing address (in Switzerland),
- plus two phone numbers - twice; but you failed to include an Email address,
- so I can't reply directly (the return address was some news server).
-
- We will install a new RF73 disk during the next days. And I'm
- rembering there are some parameters like clustersize and so on wich
- affect the performance of disks. But unfortunatly I don't remeber were
- I saw it. I tried the manuals, but its a hard work to check all.
- Perhaps some kindly soul has it handy and may help me.
-
- Our konfiguration is as follows:
-
- MVS3300 and MV3400 Dualhost over DSSI, all Disks on DSSI
- 1 RF71 System
- 1 RF71 Some Applications (Mailer and Network SW)
- 1 RF72 VMS Userfiles 16'000 files in 800 directories
- 1 RF73 MSA Files 55'000 files in 6'000 directories
-
- MSA has always a lot of files open
-
- We are tigth on the following resources: IO and CPU, we have plenty of
- memory (each MV 44 has Mbytes) and enogh diskcapacity (50% used).
-
- There are relatively few parameters you can adjust, and of those almost none
- have anything to do with performance.
-
- Clustersize is, indeed, one of the few. The tradeoff is simple: Larger
- clustersizes produce better performance (files that have larger contiguous
- pieces) at the cost of disk space. Since file allocations are always rounded
- up to the nearest multiple of the clustersize, if the cluster size is C,
- the expected space wasted per file is (0+1+...+(C-1))/C = (C-1)/2. On the
- other hand, with this clustersize, every C blocks of the file are always
- contiguous and can be read quickly. You really have to look at the expected
- usage of the disk to make a rational choice of clustersize. If the disk will
- contain a very large number of small files, it makes little sense to choose a
- large clustersize: Many of the files will be smaller than the clustersize and
- would have been contiguous even with a smaller clustersize, and even on a very
- fragmented disk it's usually possible to find a couple of adjacent clusters
- for the slightly larger files. On the other hand, if the disk will contain
- relatively fewer large files, the space lost due to rounding won't be signifi-
- cant but the additional contiguousnous will help.
-
- The only other consideration in choosing the clustersize is that you should,
- if possible, choose a clustersize that evenly divides the number of blocks
- on a track; or failing that, on a cylinder. This ensures that clusters don't
- cross track (cylinder) boundaries. I don't know the relevant numbers for an
- RF73, but they'll be in the specs. Make sure you use the number of blocks
- that will be USED; there will be some spare ones for replacing bad blocks.
-
- You can move the index file from its default position in the center of the
- disk by using the /INDEX qualifier of INITIALIZE. However, the default is
- the position that gives the best performance in normal usage; moving it
- elsewhere is only useful when you are formatting a disk that is intended to
- contain a really huge file (larger than half the disk),
-
- The /WINDOWS qualifier lets you change the number of file mapping pointers to
- be used when the disk is mounted. Larger values can give you better perfor-
- mance at the cost of memory - the window is effectively a cache - but the
- value you specify here is only a default; the MOUNT command can override it.
- So it's something you can always change later.
-
- Specifying /NOHIGHWATER will have a minimal effect on performance unless your
- use of the disk is fairly unusual. (Only if you have programs that frequently
- write to blocks of a file well beyond the current end of the file - i.e.,
- programs that leave unwritten "holes" in a file - will you be likely to see an
- effect.)
-
- Most of the remaining parameters you can set have to do with pre-allocation
- of objects for anticipated later use. For example, the /DIRECTORIES qualifier
- sets the size of the MFD - useful if you intend to create a large number of
- top-level directories. /MAXIMUM_FILES and /HEADERS control the maximum number
- of files you will ever be able to create on the disk, and the number of file
- headers to pre-allocate.
-
- Beyond this, you can increase various cache sizes through SYSGEN parameters
- like ACP_MAPCACHE. These will apply to all disks in the system, and don't
- require any change to the disks themselves.
- -- Jerry
-
-