| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1012.1 | Try this | KCBBQ::DUNCAN | Gerry Duncan @KCO 452-3445 | Sun Oct 20 1991 04:36 | 29 |
| Try 424.8 where I describe how redo logging works. Remember that
Oracle's redo logging is kinda' like Rdb's AIJ and kinda' like Rdb's
RUJ. I may have mentioned it in 646.1 as well.
Redo log entries are written to disk at EVERY commit. As I understand it,
your program does not get control until redo records are complete. As
a result, redo files can be very hot.
I'm going to give Oracle the benefit of the doubt (I'm getting soft
in my old age !) by guessing that if, during the commit sequence,
the redo disk failed, Oracle would shutdown gracefully and flush
"clean" commited database pages to disk, thus avoiding a corrupt
database. So you should only lose the transaction in flight.
What you're probably referring to when you mention "pages" getting flushed
to disk, is Oracle database pages. Redo logging and, if running, archive
logging pretty much keep this from being an issue ... just as Rdb now does
in v4.1 with undo/redo logging via the AIJ.
Just a note on Oracle redo logging. Yes, shadowing would help but
on an active and/or large database where recovery would take too
much time, I would expect the dba to have multiple redo logs
on different disks in addition to shadowing.
Finally, I don't believe any checkpoints are required to guarantee
database recovery. However, it only takes one checkpoint to
flush the unwritten, commited buffers from memory to disk.
-- gerry
|
| 1012.2 | ex | NOVA::FEENAN | Jay Feenan, Rdb/VMS engineering | Wed Oct 23 1991 19:36 | 4 |
| it is necessary to shadow the disk where the redo log resides!
-Jay
|
| 1012.3 | REDO without SHADOWING = DISASTER ? | TRCOA::MCMULLEN | Ken McMullen | Wed Oct 23 1991 20:06 | 12 |
| Jay,
I know that Phil B. said you had to do shadow redo logs in his course
the other week, but Phil is the expert on theory. I find it hard to
believe that even Oracle would not tell their customers of the
potential for database corruption if the redo log is not shadowed. But
I guess some platforms do not offer shadowing.
Are there no implementations of redo logging that would not require
shadowing of the redo log disk (including our products)?
Ken
|
| 1012.4 | | NOVA::NOVA::R_ANDERSON | My timing is Digital. | Thu Oct 24 1991 12:54 | 16 |
| From all the real-world reports I have heard, ORACLE does indeed require that
the undo/redo log be shadowed. This is required to avoid the single-point-of-
failure test that is part of the TPC auditing requirements.
ORACLE probably does not mention this strenuously in their technical
documentation because, for the majority of the ORACLE customers, losing the last
transaction does not constitute a tragedy. For those customers that require
that their database be absolutely, positively recoverable, they HAD BETTER
shadow the logs (small price to pay for some piece-of-mind).
Rdb/VMS does indeed provide single-point-of-failure protection WITHOUT
requiring that the AIJ file be shadowed. To the best of my knowledge, this
is the only database that does provide non-shadowed single-point-of-failure
protection.
Rick
|
| 1012.5 | | NOVA::FEENAN | Jay Feenan, Rdb/VMS engineering | Thu Oct 24 1991 17:29 | 8 |
| re:.3
The reason I said it is because I know that it is required for the d/c
benchmark...for Oracle. V4.1 will have a competative advantage in this
area when it is released.
-Jay
|
| 1012.6 | Dual-Log | COOKIE::BERENSON | Lex mala, lex nulla | Tue Oct 29 1991 18:43 | 3 |
| A product could certainly avoid shadowing if it implemented its own
dual-logging feature. I'm surprised they don't (for implementations that
don't offer the shadowing option).
|
| 1012.7 | Oracle uses software shadowing... | NOVA::NOVA::R_ANDERSON | My timing is Digital. | Tue Oct 29 1991 23:12 | 7 |
| In talking to the Oracle TPC-B Porting Manager at another (unnamed)
computer vendor, Oracle does indeed provide an undo/redo log shadowing
option (done in Oracle software). According to the manager, the
implementation is extremely weak and Oracle is working on replacing it
with a better implementation.
Rick
|
| 1012.8 | ?New feature | HGOVC::DEANGELIS | Momuntai | Wed Oct 30 1991 07:03 | 10 |
| � <<< Note 1012.7 by NOVA::NOVA::R_ANDERSON "My timing is Digital." >>>
� -< Oracle uses software shadowing... >-
Rick,
What version is this feature in? I've read the V6.0 docs and it only discusses
the fact that you can have > 2 redo logs (min 2), but only one is active at
any point in time.
John.
|
| 1012.9 | | NOVA::NOVA::R_ANDERSON | My timing is Digital. | Wed Oct 30 1991 20:04 | 6 |
| I don't have that information - I would assume it is the latest available
product version commercially available. However, knowing the way Oracle does
their benchmarks, it's probably a "shoe-horn" feature (undocumented) needed to
audit their TPC-B results :-)
Rick
|
| 1012.10 | | NOVA::NOVA::R_ANDERSON | My timing is Digital. | Fri Nov 01 1991 12:30 | 17 |
| Here's the scoop on the ORACLE software shadowing - as I suspected, it's not
yet a "productized" feature...
>
> BTW: Oracle claims to have a software-shadowing mechanism. Do you have any
> information on this feature? Seems unnecessary since most machines provide
> hardware shadowing...
>
Yes, the Oracle people in Sydney have supposedly setup a shadow system for
the defence department but it involved kludging the compiled code and they
wouldn't say how it was done.
The standard response is wait for version 7.0 which hopefully will hit beta
sites in the near future (which means oz will get in '93).
Rick
|