Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!seismo!rutgers!ames!ucbcad!ucbvax!cbosgd!cwruecmp!nitrex!rbl
From: rbl@nitrex.UUCP ( Dr. Robin Lake )
Newsgroups: comp.sys.mac
Subject: Re: Why is A/UX disk I/O only 50KB/sec?
Message-ID: <498@nitrex.UUCP>
Date: Thu, 30-Jul-87 10:02:13 EDT
Article-I.D.: nitrex.498
Posted: Thu Jul 30 10:02:13 1987
Date-Received: Sat, 1-Aug-87 16:13:55 EDT
References: <2363@ames.arpa> <344@swanee.OZ> <3241@pwcs.StPaul.GOV>
Reply-To: rbl@nitrex.UUCP ( Dr. Robin Lake )
Organization: The Standard Oil Co., Cleveland
Lines: 34

In article <3241@pwcs.StPaul.GOV> dennisg@pwcs.StPaul.GOV (Dennis Grittner) writes:
>In article <344@swanee.OZ> gustav@swanee.OZ (Gustav) writes:
>>In article <2363@ames.arpa>, jaw@ames.arpa (James A. Woods) writes:
>>> During Phoenix USENIX, at the Unisoft booth, on the Mac II,
>>> a "time cat /usr/dict/words > /dev/null" took about four seconds
>>> of clock time. 
>>
>>Well, let me compare this to our VAX 11/750 running 4.2BSD. The disk
>>we use is RA81:
> Lots of stuff removed about times, etc.
>
>
>>
>What you are witnessing on the 11/750 is called DEC never could
>do I/O. ...

In spite of the I/O rate deficiencies of the Unibus, consider the
overhead of the device drivers and the entire I/O strategy of the
operating system.  In the mid-1970's, we did UNIX performance evaluation
on a Unibus machine (PDP-11/45) using a solid-state disk versus a
head-per-track fixed head disk and a moving head disk.  Even though the
transfer rate of the solid-state disk is 17,000 times faster (zero
latency) than the fixed head disk, the device drivers consumed  the vast
majority of the transfer time.

With DMA transfers, only a few words need be transfered to the device
controller to initiate a transfer.  However, many thousands of instructions
need be executed before those few words are determined!

Rob Lake
decvax!cwruecmp!nitrex!rbl
ihnp4!cbosgd!cwruecmp!nitrex!rbl

Disclaimer:  This work was done prior to joining my present employer.