Xref: utzoo comp.databases:3791 comp.arch:11642
Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!gem.mps.ohio-state.edu!apple!agate!shelby!unix!quintus!jbeard
From: jbeard@quintus.UUCP (Jeff Beard)
Newsgroups: comp.databases,comp.arch
Subject: re: *big iron*, io bandwidth
Message-ID: <1268@quintus.UUCP>
Date: 4 Oct 89 16:16:41 GMT
Reply-To: jbeard@quintus.UUCP (Jeff Beard)
Distribution: usa
Organization: Quintus Computer Systems, Inc: Mountain View, Ca
Lines: 16

To avoid i/o wait time and to max(transaction rates), the ACP 
(Airlines Control Program) of prior years used interesting file layouts.

Of a disk farm of N disks, (assuming M tracks/cylinder each) the sequence of
sequential records R1,R2,...Rn were distributed to 
cylinder 0, track 1, record 1, (disk1, disk2, ... diskN). 
The Nth+1 record started at cylinder 0, track 2, record 1 and
the 2N+1  was at cylinder 0, track 3, record 1 and so on.

Thus after initaily seeking N arms (asynchronously), 
N*M records could be transferred without arm movement.  
By the wonders of S/360 channnel programming, the 
new arm positioning for the next round was restarted when the
last record from disk 1 track M was transferred, while disk 2 track M data
transfer was in progress.  Thus continuous, asynchronous data transfer 
required only rotational+transfer time.