Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!uunet!sobmips!roe
From: roe@sobmips.UUCP (r.peterson)
Newsgroups: comp.sys.ncr
Subject: Re: Copying from one tape to another tape
Message-ID: <1990Jan13.212911.14107@sobmips.UUCP>
Date: 13 Jan 90 21:29:11 GMT
References: <1892@sauron.Columbia.NCR.COM>
Distribution: usa
Organization: Sobeco Group - Montreal, Canada
Lines: 19

From article <1892@sauron.Columbia.NCR.COM>, by wescott@Columbia.NCR.COM (Mike Wescott):
> 
> It is a function of using an Intel bus (mutlibus I) with a 
> big-endian processor (MC68K).  The newer controllers (SCSI) reswap
> the bytes so tapes and other peripherals look right to application
> software.

With an amazing reduction in real I/O speed.  Say you are writing 32K
blocks - the driver has to byte-swab the 32k once before it writes the
data, and then again after the write - since the buffer is in user space,
it has to put your data back the way it came.

This caused the tape driver (on a 600 running 01.03.02) to max out at about
100K per second.  Removing the swab in both directions produces about
220K/sec - more than twice as fast.
-- 
One makes strong assumptions delving	       Roe Peterson
into the beginning of the universe...	       {uunet,mcgill-vision}!sobeco!roe
	- Stephen Hawking, Cambridge