berger@datacube.UUCP (11/16/87)
We are running osx 4.0 on a 90x (with no data cache). We seem to be
experience real performance problems primarily with the ethernet.
We mostly have a network of Sun 3's with 2 Sun's as servers. There is
cross NFS mounting between the Sun's and the Pyramid. The sun's are running
Sun OS 3.4.
The biggest problem is when a user has a home directory on the Pyramid
and is logged into a sun, even if they are working on a Sun filesystem,
they get error messages such as:
NFS getattr failed for server datacube: RPC: (unknown error code)
NFS server datacube not responding still trying
NFS write failed for server datacube: RPC: Timed out
NFS write error 60 on host datacube
NFS getattr failed for server datacube: RPC: Timed out
NFS lookup failed for server datacube: RPC: Timed out
NFS readdir failed for server datacube: RPC: Timed out
NFS lookup failed for server datacube: RPC: Unable to receive
These messages happen pretty regularly if the sun is dealing with
a file on the pyramid file system via NFS or if the sun user's home
directory is on a pyramid file system via NFS.
The biggest problem is that some programs (in particular FRAME a WYSIWYG
editor on the Sun) will hang during a write, jamming the whole Sun
workstation at times.
Any suggestions?
Bob Berger
Datacube Inc. Systems / Software Group 4 Dearborn Rd. Peabody, Ma 01960
VOICE: 617-535-6644; FAX: (617) 535-5643; TWX: (710) 347-0125
UUCP: berger@datacube.COM, rutgers!datacube!berger, ihnp4!datacube!berger
{cbosgd,cuae2,mit-eddie}!mirror!datacube!bergersteve@BRILLIG.UMD.EDU (Steve D. Miller) (11/17/87)
The Suns send out huge (ballpark 4K) UDP packets when doing NFS reads and writes. These packets get fragmented at the IP level into three back-to-back large IP packets; if the Pyramid Ethernet controller or software is slow, it could end up dropping one or more of those IP packets. This can be controlled by using the rsize and wsize options when you do a cross-mount. (Setting these to 2K should help.) There should be something in the Sun mount(8) man entry about this, and I would hope that the option appears on the Pyramid side, too. This same sort of thing happens between Sun-2s with 3Com Ethernet cards and Sun-3s. Fiddling this stuff might not help, but if you haven't tried, you should. Good luck. -Steve Spoken: Steve Miller Domain: steve@mimsy.umd.edu UUCP: uunet!mimsy!steve Phone: +1-301-454-1516 USPS: UMIACS, Univ. of Maryland, College Park, MD 20742
bob@aargh.cis.ohio-state.edu (Bob Sutterfield) (11/18/87)
In article <122700002@datacube> berger@datacube.UUCP writes: > >We are running osx 4.0 on a 90x (with no data cache). We seem to be >experience real performance problems primarily with the ethernet. >... >The biggest problem is when a user has a home directory on the Pyramid >and is logged into a sun, even if they are working on a Sun filesystem, >they get error messages such as: > >NFS server datacube not responding still trying >NFS write failed for server datacube: RPC: Timed out >... Sounds pretty familiar. We had such problems all spring and half the summer. Traced our Ethernet backbone (supecting length/late collision problems) and cleaned it up a lot in the process. Spent a lot of time watching packets fly with a LANalyzer. Really thought it was our problem. All the users were getting pretty irate at us, thinking the system was crashing and bouncing 20 times a day. Then one day, while talking to RTOC about another problem, I happened to notice that once again, all my Pyramid-based X clients' connections to my Sun had timed out and winked away, and once again swore pretty vehemently at our network. The RTOC person said "Oh yeah, that's a known problem. You see it? OK, we'll ship you a PTF and fix you right up." Sure enough, next morning I installed their PTF, rebooted, and we haven't had such problems since. So, I'd suggest you describe your problems to RTOC, because they just might be able to do something about it for you. While you have them on the phone, suggest (as I have) that they change their fix distribution policy so as to send out PTFs that solve problems of some designated severity level, to all users of that software, on a regular basis. This would seem a better idea than customers wasting a couple of months fooling around trying to solve major problems, when all along it was a known bug with a known fix. As I said, I have mentioned this to RTOC, our local sales rep, our local tech support person, and anybody at any levels inside Pyramid who would listen. Maybe if we get more votes, it mike help. "And if twenty people walk in singing the first few bars... they'll think it's a movement!" -=- Bob Sutterfield, Department of Computer and Information Science The Ohio State University; 2036 Neil Ave. Columbus OH USA 43210-1277 bob@ohio-state.{arpa,csnet} or ...!cbosgd!osu-cis!bob soon: bob@cis.ohio-state.edu