T.R | Title | User | Personal Name | Date | Lines |
---|
1870.1 | | BACHUS::DEVOS | Manu Devos DEC/SI Brussels 856-7539 | Fri Feb 07 1997 09:33 | 8 |
| Hi,
My personal and humble opinion is : Never.
The SCSI extenders are indeed not able to distinguish between a broken
SCSI and a partition.
Regards, Manu.
|
1870.2 | more infor? | USCTR1::ASCHER | Dave Ascher | Sat Feb 08 1997 09:50 | 18 |
|
Manu,
Could you please clarify? If you have a config like
>>>SCSI >>>>FIBER >>>> SCSI >>>>>>>
| | |
nodeA | nodeB
SCSI device
what would be the issue? I can imagine that if some disks
were on the left side of the fiber in the same place as nodeA
and others were on the right side there is a potential issue
- but that should be an illegal configuration rather than a
show stopper.
d
|
1870.3 | | KITCHE::schott | Eric R. Schott USG Product Management | Sat Feb 08 1997 11:42 | 15 |
| The issue is that when a normal scsi bus fails, both sides an error.
When the extender fails, both sides may not see the
same failure (or even an error except for timeout)...
In short, the extender changes some of the properties of the scsi
bus, that will cause problems with the detection mechanisms in
ASE. Thus it will not be supported.
If you are looking for DT solutions, see
http://www-unix.zk3.dec.com/www/clusters.html under Disaster Tolerance.
regards
Eric
|
1870.4 | Eric is close | FIEVEL::FILGATE | Bruce Filgate SHR3-2/W4 237-6452 | Sun Feb 16 1997 12:55 | 22 |
|
SCSI bus' can not tell the difference between a broken
bus and a partition. ASE clusters solved this problem by also
keeping track of another communication path, network.
While ASE remains in a small computer room, there is a small
exposure to a double communication error, SCSI plus network,
being evaluated incorrectly and generating a partition that
destroys the customer data. The probability is non-zero,
but accepted as being small enough by the engineers responsible
for ASE design.
If ASE were to be extended beyond the computer room, this small
probability grows as the interconnect distance increases. The
engineering that was done for computer room hazard protection,
is entirely inadequate for the longer distances that FC enables.
Until the design engineers in ASE INVENT a solution to this large
area partition issue, there will not be a fully supported ASE that
won't corrupt customer data with a long haul FC.
Bruce
|