From wiedeman@altair.acs.uci.edu  Fri May  7 00:53:26 1993
To: chemistry@ccl.net
Subject: G92 and NFS
Date: Fri, 07 May 93 07:53:26 -0700
Message-Id: <599.736786406@altair.acs.uci.edu>
From: Lyle Wiedeman <wiedeman@uci.edu>



Gee, I certainly got all the discussion I could have wanted, and more!

I've seen a range of opinions:

    * G92 wasn't meant to work through NFS
    * You'd be an idiot to use NFS; it's cheaper to go DIRECT
    * Some of us have no choice but to use NFS
    * G90 had bugs, but G92 is fixed
    * Convex G92 must be at fault, because the Sun works with the Indigo

#3 is where we stand: we have G92 only one place on campus, a central
time-sharing machine which is reasonably fast, had LOTS of core
memory, and (for most applications) a large amount of disk space.
Unfortunatly, with G92 creating files of 1GB and up, "lots" of disk
space is never enough on a shared machine.  Worse, researchers here
generally cannot afford to pay for 1GB of ongoing disk space on the
central machine; it is MUCH more economical for them to get 1, 2, or
more GB of cheap SCSI disk on their individual workstations, and
access it through NFS.  I am still curious about the performance
tradeoffs (DIRECT vs. NFS), so any reports will be appreciated.  
Also, I am sure I will hear from the Convex G92 porting team soon.

Thanks, one and all, even those who called us stupid and insane :-)

	Lyle Wiedeman                 Office of Academic Computing
	wiedeman@uci.edu              Univ. Calif. Irvine
	wiedeman@UCI.BITNET           Irvine, CA  92717
    	(714) 856-8718	    	      FAX (714) 725-2069