Re: computational model of transactions

Erwin wrote:
> > "At time T4 the $100 transaction completes, recording a balance of
> > $450." allright, but
> > at time T5 the $75 transaction aborts with a 'cannot serilize message'
> > likewise, at time T6 the $500 transaction also aborts with the same
> > message.
>
> versus
>
> >> Since each of them wants to update the very same resource (the same
> >> attribute of the same tuple of the same relvar), these transactions
> >> should be serialized anyway.
>
> > Not necessarily. Read any book on MVCC.
>
> So to summarize :
>
> If I claim those transactions should be serialized, then you say I'm
> wrong (well, the least you implied is that my statement was a bit short
> through the bend, if I read you correctly).
>
> And regarding the given example itself, you then say, as I did, that
> those transactions cannot complete normally unless they are indeed
> serialized ?

Not quite. The two transactions will be aborted in the Oracle MVCC variety. Assuming a hypothetical multiversion scheduler with timestamp ordering, only the $500 transaction will be aborted (and automatically restarted with a new timestamp). Under such scheduler, write only transactions will never conflict either.

>
> Can you please explain, because I'm afraid I don't really understand.
Received on Tue Aug 08 2006 - 05:01:36 CEST