uskmg@mathcs.emory.edu (Ken Guyton {EUCC}) (03/07/90)
Can anyone tell me about the performance of a diskless NeXT being served by another NeXT (330 Mb)? I am interested in the case where the ``diskless'' machine actually has the 40 Meg drive. How does it perform relative to a diskful NeXT (e.g., the 330 M) and to a OD-only NeXT? I would also be interested in performance relative to other systems. Any comments would be appreciated. Ken Guyton | uskmg@unix.cc.emory.edu PREFERRED Emory University | ...gatech!emoryu1!uskmg UUCP Academic Computing | uskmg@emoryu1 OR uskmg@emuvm1 NON-DOMAIN BITNET Atlanta, GA 30322 | Phone: (404) 727-7651 -- Ken Guyton | uskmg@unix.cc.emory.edu PREFERRED Emory University | {decvax,gatech}!emoryu1!uskmg UUCP Academic Computing | uskmg@emoryu1 OR uskmg@emuvm1 NON-DOMAIN BITNET Atlanta, GA 30322 | Phone: (404) 727-7651
gerrit@nova.cc.purdue.edu (Gerrit) (03/07/90)
In article <5106@emory.mathcs.emory.edu> uskmg@emory.UUCP (Ken Guyton {EUCC}) writes: >Can anyone tell me about the performance of a diskless NeXT being >served by another NeXT (330 Mb)? I am interested in the case where >the ``diskless'' machine actually has the 40 Meg drive. How does it >perform relative to a diskful NeXT (e.g., the 330 M) and to a OD-only >NeXT? I would also be interested in performance relative to other >systems. Astounding. Well, the difference is astounding. We were originally running half a dozen machines each with hard drives running 0.9 in a demo lab. We later set up 10 clients and a server in a lab and had them all running off the server with 1.0 installed. At that time, none of the machines had swap drives configured. 0.9 in the lab wasn't very well performance tuned, but the machines were faster than Sun 3/60's. Without the swap drives, the machines were more like memory-starved 3/50's. Then our swap drives came and the machines are now outrunning 3/60's again. I'm not sure if they are faster than they were when everything was running off a local hard disk, but swapping locally seems to buy you a *lot* of performance over network swapping. Keep in mind also that the NeXT servers typically have only a single disk drive with only a single disk arm. There was a *lot* of contention on that disk arm when 10 machines plus the server (a print server at that) were all busy. I think another performance enhancement I'm going to try sometime is using multiple disks to better balance disk performance (I have a couple of machines with 330 Meg drives that I can try mounting user home directories on and such). gerrit