Dear Managers,
I have some questions about sharing storage among multiple machines. We are
trying to set up a system to reduce some large astronomical datasets (~50
Gb each). We will have several independent machines (one DS20, two
XP1000's) each running different types of reduction software. Our hope is
to have a RAID Array 3000 or 7000 with all of our disks (~250 GB or more)
which is then shared, to some extent, among the three hosts using an
UltraSCSI hub. Basically this would consolidate our storage in one unit and
leave the ethernet network between the machines free for user traffic. We
would like to share the disks as much as possible among the different
machines, but I realize that mounting partitions simultaneously on two
machines won't be available until at least Tru64 v5.0 and will require a
cluster in any case (which I don't believe is possibile with the XP1000's).
We don't really need any particularly high-availability features in the
system (e.g. automatic failover) and the data reduction will not be
particularly I/O bound (i.e. it will generally take longer to process the
data in memory than to read it from disk).
Questions:
A) Is it true that the RA3000 can only be shared via an UltraSCSI Hub if it
is being used in a cluster and/or only with Windows NT?
B) If I share the "storage subsystem" in a non-clustered configuration (as
permitted by the RA7000), how flexible can I be about switching disks
between one machine and the other? Is it as simple as unmounting a
partition from one machine and mounting it on another? Or does it require
serious reconfiguration on the RAID Array?
C) Would I gain anything about going to Fibre Channel (other than the
obvious speed increase)? Or is the problem with sharing more at the OS
level? What is the status of Fibre Channel support in Tru64 Unix?
D) Are there any third party solutions that might be better suited to these
needs?
All in all, the concept of a "Storage Area Network" (SAN) has been
recieving a lot of hype (and is quite attractive), but for now its
application still seems quite limited (only in clustered configurations or
in particular database applications).
I will summarize, but thanks in advance for the advice.
kevin reardon
Osservatorio Astronomico di Capodimonte
Received on Fri Jun 04 1999 - 11:03:24 NZST