Newsgroups: comp.sys.sgi
Path: utzoo!utgpu!news-server.csri.toronto.edu!helios.physics.utoronto.ca!physics.utoronto.ca!chlebana
From: chlebana@physics.utoronto.ca (Frank Chlebana)
Subject: PIO transfer rates 25s vs 35s
Message-ID: <1991Apr29.203803.2093@helios.physics.utoronto.ca>
Sender: news@helios.physics.utoronto.ca (News Administrator)
Nntp-Posting-Host: helios.physics.utoronto.ca
Organization: University of Toronto Physics/Astronomy/CITA
Date: Mon, 29 Apr 1991 20:38:03 GMT

I am using Programmed IO to read data from an external VME crate into local
memory of the 4D/35S. The cpu executes load and store instructions from/to the
bus extender registers to/from local memory. The bus extender I am using does
not operate as a VME master, so I cannot use it as a DMAC. 

To do the transfer I use a similiar routine as bcopy(). Prior to the transfer
splhi() is set, so the machine is virtually dedicated to the transfer.

On the 4d/25s I was able to get the following transfer rates

reads     4.3 MB/s
writes    3.8 MB/s

On the 4d/35s the rates are

reads     2.6 MB/s
writes    2.8 MB/s

These are typical times for a 2047 longword transfer, where 1 MB==2**20.

Am I missing some "trick" that can be used to increase the transfer rates
when I access a VME slave? Can the 4d/35s initiate DMA from VME slaves??

Frank Chlebana
chlebana@oldkat.physics.utoronto.ca