| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 2040.1 | And the answer is??? | DYOSW5::WILDER | Does virtual reality get swapped? | Wed May 07 1997 07:10 | 8 |
| Does the lack of a response mean that pulling the mc cable off IS the
best way to simulate a memory channel failure? Has anyone done this? Is
this the BEST way in a hubless environment?
Thanks,
/jim
|
| 2040.2 | yes | AFW4::CLEMENCE | | Wed May 07 1997 07:24 | 0 |
| 2040.3 | Thanks, now, what will happen | DYOSW5::WILDER | Does virtual reality get swapped? | Thu May 08 1997 07:09 | 12 |
| Thanks.
Now, what should we expect? This will cause a partitioned database
(since we are running OPS). Therefore, I would expect one node to
crash. Is there anyway to fresee which node? Will the node running the
Director survive, or is it a coin toss? Or (heaven forbid) will both
nodes crash, and let the reboot make the decision on the survivor?
Thanks,
/jim
|
| 2040.4 | | KITCHE::schott | Eric R. Schott USG Product Management | Thu May 08 1997 12:50 | 3 |
| Hi
Do you have quorum disk setup?
|
| 2040.5 | Tie-breakers defined | DYOSW5::WILDER | Does virtual reality get swapped? | Fri May 09 1997 07:14 | 18 |
| Yes, there is a quoru/tie-breaker disk. In fact, there are 3
tie-breaker disks, one per HSZ pair.
Now, I realize that with 3 disks, the system that sees at least 2 will
be the winner. My question is, what do we expect to happen? Will the
winner force the other system to crash? Reboot? What?
If the losing system reboots, what will happen if we re-connect the
Memory Channel cable while the systems are up?
These are questions I know the customer will be asking, and I would
like to try to appear knowledgable and have these answers BEFORE they
ask.
Thanks,
/jim
|
| 2040.6 | And the result is..... | DYOSW5::WILDER | Does virtual reality get swapped? | Fri May 23 1997 09:51 | 18 |
| Well, for those of you still wondering, we had to find out what would
happen at the customer site (I love learning in front of the customer).
When the MC cable was disconnected, the node with the tie-breaker disks
survived. It forced the other node to panic. After that node crashed,
it re-booted. When TCR came up, it found the surviving node on the
network and the SCSI. However, when it discovered that it could NOT see
it over the memory channel, it displayed the error message and shgut
itself down to single user mode.
Oracle performed exactly as we had predicted, recovering transactions
that had not committed on the dead node, while the surviving node only
paused once briefly.
The customer is impressed!
/jim
|