Issues with Oracle and VMS Host Based Volume Shadowing?

[John Donovan_4]

Oh I'm sorry I misread your post. I thought you were talking about a REAL database!
Um... shhhh ...Rdb is a classic.

Thanks for the 10! :^))

[Dave Gudewicz]

Jim Geier wrote on Oct 7, 2005:

All of the relevant HP hardware and software support groups are involved.

--

Its been about a month and I wondered if there has been any updates on this issue in Orange, CA.

Dave...

[Robert Brooks_1]

I have spent a fair amount of time digging around -- it's clear that no one in VMS Engineering has not been made aware of any issue that is specific to Oracle and HBVS.

Given that, it's rather difficult for me to speculate exactly what the issue is, and what a potential solution may be.

If someone has more direct knowledge of the problem, please contact me directly at
firstname.middleinitial.lastname@hp.com so that VMS Engineering can become involved.

My middle initial is A

There are several undocumented tools that are available that ship with VMS that would allow us to gather information as to what is happening. The tool I'm specifically thinking of is the DKDRIVER logging tool, available within SDA. With this tool, we can see exactly what is going on between the controller and the shadow set member(s).

That a device has logged many errors and has not been tossed out of a shadow set is not necessarily cause for concern; we'd need to see the error logs to see the exact type of error.

In any event, please contact me at the address above if you've got more info on this
alleged issue.

-- Rob

[Robert Brooks_1]

I mistakenly wrote . . .

I have spent a fair amount of time digging around -- it's clear that no one in VMS Engineering has not been made aware of any issue that is specific to Oracle and HBVS.

---
Please excuse my inadvertent use of a double negative; I meant to write that no one in VMS Engineering has been made aware . . .

-- Rob

[Keith Parris]

I suspect the cautionary wording about support of replication solutions on the Oracle website is not related to HBVS at all, but related to potential problems with some controller-based data replication products. Databases depend on the proper ordering of writes for correct and successful recovery after failures. For example, they tend to write the journal/log file first, then update the main database area. When they do recovery, their algorithms depend on this write ordering having been preserved. EMC's SRDF reportedly replicates data by passing the contents of entire tracks at a time, which can effectively re-order writes and cause a replicated copy to have inconsistent and unreadable (or corrupted) data if the replication link gets interrupted (as it does when a disaster occurs). StorageWorks EVA-CA and XP-CA products have the option of guaranteeing the order of all writes, both during steady-state operation and during catch-up operations, so you'll never be left with an inconsistent or unreadable copy of data. More detail on these problems is in my presentation on 'Disaster-Tolerant Cluster Technology & Implementation' at http://www2.openvms.org/kparris/

Host-Based Volume Shadowing does only strictly synchronous mirroring, which prevents it from being susceptible to the write-ordering problems I discussed above, since the writes are all guaranteed to be completed in order, always (HBVS doesn't return successful status to Oracle (or any other application doing writes) until the data is successfully written to all the shadowset members.

[Dave Gudewicz]

Rob and Keith,

Thanks for the updates.

Dave...

[Rob Young_4]

Keith writes:

EMC's SRDF reportedly replicates data by passing the contents of entire tracks at a time, which can effectively re-order writes and cause a replicated copy to have inconsistent and unreadable (or corrupted) data if the replication link gets interrupted (as it does when a disaster occurs).

---

That isn't correct. I'm assuming you are
talking SRDF/A as SRDF/S is synchronous.
With SRDF/A you have:

Dependent write logic
If "A" is a predecessor and "B" is a dependent write
Any I/O "B" that arrives after I/O "A" has been acknowledged to the host, must be dependent on "A"

SRDF/A ensures that:
"A" and "B" are in the same Delta Set or
"B" is in later Delta Set

These Delta Sets (aka Cycles) of I/Os, rather than the I/Os themselves, are being ordered by SRDF/A

Symmetrix ensures that dependent write relationships are honored during Delta Set switch or Write Folding

---

My interpretation is to think of the Delta Sets as Point In Time copies. You lose
x number of transactions but you have data
consistency with SRDF/A.

Now there is a whole lot to this, I certainly
don't want to get caught up in the minutia.

Rob

[Keith Parris]

Thanks for the update/correction, Rob.

[Keith Parris]

Jim wrote:
> hundreds or thousands of uncorrectable
> errors are generated.
...
> How does one simulate such an error or
> series of errors.

If you could identify what type of errors were returned, one could use LDDRIVER to return errors. You can set up all requests to a physical disk to go through LDDRIVER, and then you can ask it to return a specific error code when a specific LBN is accessed, for example.

[melanie_17]

With regards to Dave Gudewicz's question posted in his 10/06/05 15:49 entry:

"Is VMS HBVS **officially** supported by Oracle? If no, please cite the reference."

I work closely with my Oracle Support counterparts, and after reading thru this thread, I asked specifically if there are any Metalink or internal Oracle Support documents which indicate that OVMS HBVS is not supported. The feedback I received from Oracle was as follows:

--------------------------------------------
Oracle support does not tell customers that HBVS is not supported and there are no Oracle documents indicating that this is the case.

In fact, they stated that almost everyone (ie OVMS sites) uses HBVS, and that it's considered a part of a sound hardware redundancy/contingency plan.

They only thing that Oracle definitely does NOT support is write-back cache. That is a situation where the OS indicates that the data has been written and really might not have been at that time. Write-back cache is not supported on any platform.
--------------------------------------------

If anyone has information to the contrary, please feel free to contact me, as not all of Oracle support is OVMS literate, and there are times when information is mis-communicated because of a lack of understanding of the underlying functionality of OVMS. We try to educate and inform whenever possible.

rgds,
melanie