Deferred Transaction Recoverability

Dear all,

Quoting about "deferred transaction recoverability": (pp.678-679, Jason Couchman):

<quote start>

Deferred Transaction Recoverability Defined

Oracle8i has altered transaction recovery behavior in several ways to increase the overall availability of the Oracle database during startup after instance failure. First, a feature that has already been available is the fast warmstart feature, which allows committed and uncommitted transactions executed up to the time of instance failure. This is called the roll-forward process. Thus, Oracle gives greater database availablility during startup.

Another aspect of deferred transaction recovery worth noting in Oracle8i is fast-start parallel rollback. Fast-start parallel rollback in Oracle8i enables a SMON to use parallel query slaves to complete the rollback operation. Parallel rollback is automatically started when SMON determines that the dead transaction had generated a large number of rollback blocks. The current threshold for determining init.ora parameter FAST_START_PARALLEL_ROLLBACK, which supports three possible values for this parameter: FALSE, LOW, and HIGH.

With the introduction of RMAN in Oracle8, DBAs were given the ability to back up the Oracle database in parallel. This feature of RMAN greatly reduced the time needed to execute the backup strategy. However, parallel backup did not assist DBAs in dire straits when fast recovery time was required for success. By permitting DBAs to use parallel transaction recovery, the overall speed of database recovery is increased.
<quote end>

Honestly speaking, I don't fully understand the three paragraphs although I read them over and over again. I cannot find the key phrase 'deferred transaction recoverability' in the doc, either. (I doubt it's gonna be in the test) The magic word 'parallel' can be accompanied with many things, such as 'query', 'server', 'DDL', 'DML', and now 'backup', 'rollback' and 'transaction recovery', and it always successfully causes me headaches. :(

I am still confused by the relation between datafiles, logfiles, and rollback segments. First, since the before-image of data is stored in rollback segments (datafiles too), it's enough information for recovery, why bother inventing sth. called logfiles? Isn't it like riding a horse on a running car? Second, the 'roll-forward' process executes both the committed and uncommitted transactions on instance recovery, which, I assume, means to store data in datafiles according to the information provided by logfiles. This is really wierd, aren't uncommitted transactions supposed to be rolled back on instance recovery because they are uncommitted when the previous instance failure happens? Does it also mean that committed transactions can be recorded in logfiles but not yet stored in datafiles? Third, what does it mean by 'defer'? What is 'deferred'? How can the instance recovery process be speeded up from it? Anyway, I wish the author could have explained in more details.

Your help will be highly appreciated.

Dino Received on Thu Aug 23 2001 - 09:48:44 CDT