home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!uvaarpa!darwin.sura.net!paladin.american.edu!auvm!MCMAIL.CIS.MCMASTER.CA!DFRASER
- X-Envelope-to: IBM-MAIN@RICEVM1.BITNET
- Message-ID: <Pine.2.4.9207231400.B12273@mcmail>
- Newsgroups: bit.listserv.ibm-main
- Date: Thu, 23 Jul 1992 14:18:05 -0400
- Sender: IBM Mainframe Discussion list <IBM-MAIN@RICEVM1.BITNET>
- From: Doug Fraser <dfraser@MCMAIL.CIS.MCMASTER.CA>
- Subject: Re: 3390 blksize calculation
- In-Reply-To: <199207231536.AA08846@mcmail.cis.mcmaster.ca>
- Lines: 37
-
- On Thu, 23 Jul 1992, Jerry Bryan wrote:
-
- > In article <IBM-MAIN%92072210340646@RICEVM1.RICE.EDU>, Zoltan Forray
- > <SSTSZXF@VCUVM1.BITNET> says:
- >
- > >For 3380 type disks, a half-track block (23400) is usually optimal for most
- > >things like load libraries. If you used anything much larger, you would waste
- > >a large amount of space on each track.
- >
- > Reasonable people can disagree about this, but I would have said that
- > half-track is best for most things *except* load libraries, where
- > 32K blocking is best. Load libraries are RECFM=U, which means that
- > the blocks are variable length. Many blocks are small to medium,
- > no matter what the block size is. A block which *can* fit on the
- > remainder of the track *will* be placed on the remainder of the
- > track and vice versa. Here are some examples. 1) First block is
- > small, second block is half-track (fits), third block is half-track
- > (doesn't fit, space wasted). 2) First block is small, second block
- > is 32K (fits), third block is small (fits). 3) First block is
- > 32K, second block is small (fits), third block is 32K (doesn't
- > fit). Construct the example of your choice to prove whatever you
- > want to prove. My personal humble opinion is that 32K is fine,
- > maybe even best, for load modules, because so many of the blocks
- > are small. Why not get the advantage of the occasional 32K block?
-
- We run mvs 2.2.x now. I believe that this was true for mvs 2.1.7 as well.
- I did some experiments on this. Our policy now is 32k load libraries for
- everything new. My experiments involved actually looking at the lengths
- of the blocks of the load modules on the disk.
-
- IEBCOPY now has a copymod statement which will copy and reblock load
- modules. You may have to be careful about concatenation of libraries. I
- don't remember.
-
- I recommend half track blocks for other (sequential) datasets, allocating
- in blocks, and rounding to cylinder boundaries if the dataset is over 1 or
- 2 cylinders in size.
-