Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!cbnewsj!duckie
From: duckie@cbnewsj.att.com (john.c.mc millan)
Newsgroups: comp.periphs.scsi
Subject: Re: Multiple Initiators, Single Target
Summary: Sure...why not
Keywords: Multiple-Initiator
Message-ID: <1991Apr16.154913.2067@cbnewsj.att.com>
Date: 16 Apr 91 15:49:13 GMT
References: <25@earth.tsd.arlut.utexas.edu>
Distribution: na
Organization: AT&T Bell Laboratories
Lines: 36

Sane or not, we run a cluster of 3 AT&T 3B2/700's which "share" SCSI buses.
Each UNIX system has:
    - one Single-Ended, internal SCSI bus with its core UNIX facilities:
	+ 2 or 3 disks [incl. file systems ROOT, /usr, SWAP...]
	+ 1 Exabyte tape
	+ "other tapes" -- 9 track, cartridges, etc.
    - two Differential SCSI buses shared across all three 3B2's:
	+ DS0= 3 Host Adapters + 3 dual-disk targets
	+ DS1= 3 Host Adapters + 2 dual-disk targets + 2 single-disks

We found the cluster was more reliable with the Exabytes un-shared.
The other tapes only had S-E adapters.

Re-booting can be a bit dicey: the reboot diagnostics of the 3B2
occasionally confuse the running systems regarding a shared bus.

Occasionally, we have simultaneously mounted a partition read-only
on several systems.  Typically, however, we use RFS for file system
sharing and the 3B2's may share drives but not partitions of those
drives.

If a 3B2 crashes, the other two systems can FSCK the damaged
file systems while the crashed 3B2 is receiving special attention.

When a system is taken out of service for an extended interval,
the file systems it usually directly mounts -- and makes available
to other systems via RFS -- are mounted by the other two 3B2's
and made available via RFS.  Since the disks are already cabled
in to each system, this transfer is done without taking down
the systems.

Nuff' said.  I presume this is the manner of target sharing to which
you were refering.


John McMillan -- jcm@pegasus.att.com -- muttering for self, only