Thanks to Paul Grant, Tom Blinn, and Allan Rollow for their informative
responses.
I had originally asked a number of questions:
1. Since DUNIX can not access SCSI target devices beyond ID 7, can the
host adapter be set to an ID beyond 7 to save the lower numbers for
storage devices? The real question was if it's possible to create a
shared SCSI bus with up to 8 hosts with shared access to up to 8 devices.
The short answer is no; host adapters with IDs beyond 7 is not supported
by DUNIX. However, the HSZ series of controllers, like most RAID
controllers, have a host side SCSI interface that can accept connections
from up to 7 hosts, and drive side interfaces that can each connect up to
7 devices. I don't remember if there were Wide SCSI versions of these
that will connect up to 15 devices each.
BTW, the HSZ70 looks like one real powerfull controller. It looks like
it can allow heterogeneous hosts to access different disk sets.
Paul Grant also mentioned about the SCSI hubs that appears to be able to
isolate the host side of the bus so that if a connected host crashes, it
will not bring down the whole bus. It also provides a repeating and
extension function so that cable lengths of up to 25m can be used on the
differential connections.
Tom Blinn also mentioned that DEC's direction is toward using FiberChannel
instead of UFW SCSI as the shared data bus.
2. The second question was whether TruCluster can run over a shared
SCSI bus, and the answer seems to be a limited "yes". Two hosts may
access the same set of drives on a shared SCSI bus in a static manner.
TruCluster does not have the features of the old LAVC on a shared SCSI
or DSSI bus. RAID controllers such as the HSZ70 are fully capable of
allowing multiple host accesses, but I'm not certain if TruCluster has
the distributed file and lock management that the VAX Clusters had. The
docs say TruCluster has a DLM to control access to "resources", so I
would think that file systems would be a "resource" that it can control.
Also, tightly coupling CPUs in a cluster requires the Memory Channel
interfaces, and a MC hub, which is like a HIPPI switch, if there are
more than 2 nodes. We don't need that kind of connection yet.
For our current applications, we are using a home-made file server with
very limited capabilites in a small cluster of heterogeneous computers
connected together with a Gigaswitch/FDDI. It looks like one of the
Network Storage Arrays can actually support file service over FDDI in
such a configuration. We may explore this a little more in the future,
but I think if we moved to an all-Alpha/DUNIX environment, a cluster
server such as the HSZ70 in an RA7000 may be the more efficient way to go.
Again, many thanks to those who replied.
eyc
Received on Sat Sep 19 1998 - 00:24:00 NZST