The Question is:
I want to ask more about using SCSI in building serious I/O subsystems (re:
Given the same redundancy and failover options being employed, is SCSI based
cluster I/O ( with multiple U SCSI adaptors on each node, multiple buses,
RA8000 units with redun. HSZ80s etc) reliable as CI based I/O ? Are there high
availability large cluste
rs using SCSI happily for I/O in place of CI ?
The Answer is :
There is no answer to the question, as there are a large number of
factors involved -- not the least of which involve the cost of the
host and storage hardware involved, the value of the data itself,
and the value of the accessability of the data. These factors are
trade-offs, rather obviously.
SCSI is not used as a cluster communications interconnection, SCSI is
solely permitted and solely used as a storage interconnection. (One or
more cluster communications interconnects is required when a multi-host
SCSI configuration is planned.)
SCSI is a single-channel path, while CI provides two parallel paths.
Disconnection of SCSI bus requires controller support (to quiesce the
SCSI bus) where upgrades to hardware on CI cabling is rather easier.
Recent OpenVMS versions (V7.2-1 and later) support multi-path failover
of SCSI widgets with specific HSZ-series controllers -- configurations
where specific HSZ controllers are connected across two SCSI buses.
Older OpenVMS releases and older HSZ controllers permit connections
to only one SCSI bus, meaning that any bus fault (SCSI controller or
SCSI widget) can trigger more general failures.
Host-based volume shadowing can be of interest.
High-end storage configurations typically include Fibre Channel widgets.
Other topics of interest include 1354, 1727, 1766 and 4348.