Path: utzoo!utgpu!news-server.csri.toronto.edu!mailrus!cs.utexas.edu!usc!wuarchive!psuvax1!psuvm!pmw1 From: PMW1@psuvm.psu.edu (Peter Weiss) Newsgroups: comp.sys.ibm.pc,psu.micro Subject: Re: Cluster Size on disks? Message-ID: <90098.170841PMW1@psuvm.psu.edu> Date: 8 Apr 90 21:08:41 GMT References: <90096.100956KDA101@psuvm.psu.edu> <90096.104817ESH101@psuvm.psu.edu> <90098.164830KDA101@psuvm.psu.edu> Organization: Penn State University Lines: 37 In article <90098.164830KDA101@psuvm.psu.edu>, Keith "Petto" Alexander says: > >In article <90096.104817ESH101@psuvm.psu.edu>, Eric Hvozda > says: >> >>My question what is a cluster? Is is equated with the BLKSIZE on mainframes, >>or is it some kind of I/O buffer size? > >A cluster is the smallest amout of space you can alocate to a file. On the >mainframe here at PSU the smallest amout of space is called Block size as >you have noted, if a file gets larger than one block by even one byte, it >takes up two blocks. A BLKSIZE in the MVS world has no direct relationship to disk/tape geometry, but is often times used as the unit of physical data transfer in Queued non-VSAM access methods. (Typically, VSAM does it's physical I/O based on Cluster Index.) The basis of queued record processing is the LRECL, which populates the BLKSIZE. Nevertheless, chosing a _good_ BLKSIZE can improved the performance of your program (sometimes, though, at the expense of system thruput in a multitasking system). On a real 3380 DASD controlled by MVS, the minimum unit of allocation is the track (47,476 bytes). Though the Channel Command Word has 16 bits for a block length (65,535 bytes), I don't recall that the standard access methods support more than 15 bits (32,767 bytes). Consequently, based on the CKD geometry, it would appear that a maximum BLKSIZE of 23,476 bytes, would give a 98.9% utilization of the track (Source - IBM 3380 Direct Access Storage: Reference Card form GX25-1678-4 (October, 1987)) assuming non-keyed data. I don't know what happens to track utilization when you start emulating another device type on a foreign geometry. -- Peter M. Weiss | (this line intentionally left blank) 31 Shields Bldg (the AIS people) | advertize here, reach Mega populi University Park, PA USA 16802 | Disclaimer -* +* applies herein