home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!ames!saimiri.primate.wisc.edu!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!ucbvax!lrw.com!leichter
- From: leichter@lrw.com (Jerry Leichter)
- Newsgroups: comp.os.vms
- Subject: re: Re: Bound volume sets: are they a bad idea?
- Message-ID: <9211141419.AA13439@uu3.psi.com>
- Date: 14 Nov 92 13:18:25 GMT
- Sender: daemon@ucbvax.BERKELEY.EDU
- Distribution: world
- Organization: The Internet
- Lines: 65
-
-
- [An earlier article mentioned ... "considering the idea of taking two
- 600Mb disks and binding them into a 1.2Gb volume set." It continues:
-
- "As far as I can reason it out, the pro is some degree of load
- balancing between the disks, and no management hassles sorting out
- what to put where. The con is a halved MTBF on the bound set. Backup
- shouldn't be a problem, both together will fit on one DAT."
-
- Mr. Brydon responds:]
-
- Several other posters have mentioned the MTBF aspect. Also, keep in
- mind that you are taking 2 'large' volumes and making them into one.
- This effectively increases the size of the cluster factor. If it is 4
- on one of these disks, it will be 8 on the volume set
-
- Eh? What makes you say this.
-
- (Same with a
- stripe set, too).
-
- I quite agree that it MIGHT happen on a stripe set.
-
- In theory, you can make the clustersize anything you like. However, there is
- a limitation, the deeper origins of which escape me a the moment, but they
- manifest themselves as requiring that BITMAP.SYS be no more than 255 blocks
- long. BITMAP.SYS contains one bit for ever cluster on the disk; a bit is
- set if the cluster is in use. At 8 bits per byte, 512 bytes per block, and
- 255 maximum blocks, there can be at most 1,044,480 clusters on the disk.
- A million blocks is roughly a 500Mb disk; no larger disk can have a cluster-
- size of 1. A 600Mb disk must have a clustersize of at least 2; a 1.2Gb disk
- at least 3. (These numbers might be one higher because a number like 600Mb
- is not exact - is it really 605*2^20, or what? The actual formula for the
- minimum cluster size is just blocks/1044480, rounded UP. Note, BTW, that
- because of the rounding up, it is not in general true that the minimum
- clustersize double when you double the number of blocks - it can be one less.
- This is a significant effect when the clustersizes involved are small.)
-
- Members of a bound volume set retain their own "block address spaces" and
- their own bitmaps. Binding volumes doesn't change the limit on the cluster-
- size.
-
- A stripe set, on the other hand, looks like one big disk. There is just one
- copy of the bitmap which must cover the entire set. Hence, binding two
- 600Mb disks into a 1.2Gb stripe set would, indeed, increase the minimum
- supportable clustersize.
-
- So if you have a lot of files that are small (ummm,
- in fact, even if you don't), you will be wasting an average of 4
- blocks per file. Even if this is not a concern in terms of disk
- space, remember that disk read/write requests are done in clusters,
- not in blocks. If your application wants one block from disk, it
- reads in the whole cluster.
-
- This is just plain false. While it's true that, IF you want to read ahead,
- reading the rest of the cluster is a good idea, there is nothing in the
- disk drivers and certainly not in the hardware which cares about cluster
- boundaries. In fact, this file-system-level abstraction doesn't even exist
- at this level.
-
- Even if most of the cluster is 'unused'
- space on the disk.
-
- -- Jerry
-
-