Performance of controllers doing tr

[neese at adaptex.UUCP]

I don't know about the WD algorithm for read aheads, but all of our
controllers that have read ahead, except for the ACB-2322D, will abort
read ahead on the receipt of a command.  The abort takes place after
the current sector read is completed.  On a drive with 32 sectors/
track the maximum time a command will be held off is 520 micro-seconds.
On a drive with 26 sectors/track, the maimum hold off time will be
640 micro-seconds.
On a note about the comparisions, one comparison was done with ENIX and
the other ISC 2.0.2.  While I find the numbers useful and interesting, it
should be noted that there is a dramatic difference between the filesystems
of these two operating systems.  ISC 2.0.2 has a fast filesystem, which
means that larger multi-sector requests will be made, as compared to ENIX
which uses the traditional SYSV filesystem.  These requests will typically
be 1K in size.  
I will soon have a system here, that I will be able to benchmark under
ISC 2.0.2 and SCO 2.3/3.2UNIX.  I will test the ACB-2372C (RLL)/ACB-2322B
(10Mbit ESDI)/ACB-2322B-8 (15MBit ESDI)/AHA-154[02]A (SCSI).  I plan to test
the controllers with like drives (in physical design i.e. same number of
heads/sectors per track) in like environments (i.e. same buffer cache size,
clean filesystems).
  While I do not beleive that any benchmark is accurate in determining the
real performance in a multi-user environment, the numbers should be good
for static comparisons.
  This will take a while to get done, but when done I will post the results
and the benchmarks (or code) that I used and the environment and settings
for the buffer cache.


			Roy Neese
			Adaptec Central Field Applications Engineer
			UUCP @ {texbell,attctc}!cpe!adaptex!neese
				merch!adaptex!neese