Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!sdd.hp.com!hplabs!hpfcso!hpfcdc!perry From: perry@hpfcdc.HP.COM (Perry Scott) Newsgroups: comp.sys.hp Subject: Re: Mixed 700/300 clusters and local disks Message-ID: <5570650@hpfcdc.HP.COM> Date: 29 May 91 23:02:05 GMT References: <1991May28.190640.5789@sunee.waterloo.edu> Organization: HP Fort Collins, Co. Lines: 87 >Hello, I was wondering if someone in the HP labs could take some time >and ponder a few bizarre questions. Actually, these are good questions. You are correct in stating that the cs80 driver is in the 8.05 release, but does not support disk drives. There are two ways I see around this. Option 1 calls for running an unsupported configuration - hang all your disks off the 700 EISA HPIB. This is likely the performance winner, but I don't know if you like running bleeding edge software. We've tried it here in the Lab, and it seems to work. The only reason it is not supported is that our poor driver writer (too busy to read notes :-) hasn't had time to do anything other than the "MUST" list for cs80 peripherals. The cs80 driver is exceedingly complex, and early versions were buggy. The driver has firmed up quite nicely in the past couple of months. The official support policy for unsupported configurations is that if you run into a problem, you're on your own. The request may be forwarded to the Lab as an enhancement request, but there is no guarantee that it will be fixed. Option 2 is supported configuration, and I am currently running my 350 workstation on it. This calls for leaving the HPIB disk drives on the 300 server, making it a node for a 700 server, and using local mounts (a new 8.0x feature). There are various I/O things that are unsupported on cnodes, but they are minor problems. The only weird thing I've seen so far is that you can turn a 350 into a 310 by doing a "find" from the 700 on a disk mounted on the 350. The 350 spends most of its time trying to keep up with the Snake. No contest. The 350 slows way down. With that said, I'll address some specifics... >I have been working with a number of SE's for a while >now, trying to come up with different options but it always seems to >come down to nothing really works. It's good you're working with your SE's. If the SE can't answer it, they call the factory. That's how we find the corner cases that slipped through our "what if ?" filter. >I can not hook up >my HPIB DAT tape to it which is really causing me a problem since that >is how HP is distributing software for the 700's. Ouch. No solution here. Maybe you can rent a SCSI DAT or CDROM until we support what you have. >Also I have been told that I might, or might not be able to hook up >our 795?B HPIB disks to it. Unsupported. Supply your own bandaids and backup the disk nightly. >The biggest unknown are the disks. I have been told that under v8.0, "local" >disks are supported on the diskless clients. However, I have not been able >to ascertain exactly what the capabilities of these local disks are. You connect the disk to the workstation, and mount it on the filesystem. The semantics are the same as if the disk were connected to the root server. Everyone in the cluster sees the disks. They can be NFS mounted by other systems. There are some exceptions, like not being able to mount on another workstation's local mount (a reliability issue). The other gotcha that I mentioned is that the 300 can't begin to keep up with disk requests from the 700. I've also noticed that two 700s talking to each other can pretty much consume a 10 Mbit LAN. >Anyway I hope someone can answer these because I am just about at my wits end >this. As near as I can tell, the 700's just don't want to talk to HPIB >devices, even with the HPIB card. The supported HPIB devices (sorry, I don't have the support matrix on my desk) are the 9144 tape, some printers, and selected plotters. >Gordon As always, I don't speak to the Net in an official capacity for HP. Hope this helps your planning. Perry Scott HP Ft Collins