Re: atomic commit;begin for long running transactions , in combination with savepoint.

my understanding was it used more resources than read committed
because it keeps track of the version id of rows selected
so far in a transaction, "transaction-level consistency",
so it has the potential to do the xmin co-selecting , and checking,
if it were a transaction isolation level in postgres.
google found my reference, and the reference mentioned it was
different from serializable.

On Mon Oct 15  9:09 , "Trevor Talbot"  sent:

>On 10/15/07, Syan Tan  wrote:
>
>> >Also keep in mind that MVCC is not the only way to implement
>> >transactions; pure locking is more common in other databases.  In the
>> >locking model, most transactions prevent others from writing until
>> >after they are finished.  Rows simply can't have different versions
>> >(and of course concurrent performance is awful).
>>
>> what about postgresql doing something like snapshot isolation level as per
>> the enemy M$ ?
>
>SQL Server is normally a pure locking database; from what I can tell,
>its snapshot isolation level adds a limited form of MVCC above that,
>making its concurrent behavior closer to PostgreSQL's:
>http://msdn2.microsoft.com/en-us/library/ms345124\(d=printer\).aspx