Hello, plinehan!

Wednesday, October 19, 2005, 6:47:15 PM, you wrote:

>> Need some correction. InterBase's multiversion
>> architecture is named as Multi-Generational Architecture
>> (MGA). So, most of documents about Interbase refer to
>> MGA, not MVCC.

p> Aw, come on! That's just semantics, whether we
p> call it MVCC, MGA or record-shadowing or
p> record-versioning or Row Versioning. 8-)

Look at another point. Someone who invented
MVCC talks - "we support MVCC, other support...
but does FB supports MVCC? No, it supports
something named MGA. But we invented
MVCC, are we not the best?".

p> I've just thought of a nice definition -
p> record-piggybacking. Any takers?

:-)

p> I think it's ORA-1555 (or something like that) is
p> the error if your undo log gets too big. Their
p> architecture does allow for PITR though.

Yep.

p> I think they had a read version of MGA since
p> version 4 but the whole system only became
p> MGA since 7 - but I can't remember where I
p> read this - have been reading a lot of
p> different sites about this.

Really, as I understand, this wasn't MGA at
all. Until version 7.2 Oracle supported
snapshot transactions by copying all data
from tables being read. And only then they
implemented record versions, AFAIK.

>> - PostgreSQL have MVCC since ... (don't know).

p> Here
p> http://pgsqld.active-venture.com/history.html

p> It seems to imply that there used to be
p> a locking mechanism but that it was
p> replaced in and around 1995, but it's
p> not that clear.

Yes, nearly 1995.

>> - MS SQL 2005 will have versioning same as in
>> InterBase/Firebird.

p> Will be similar (if ever realeased) as Yukon.
p> Many went there, most died poor and cold.

No, I read article about MS SQL 2005 MGA last year,
and was at MS SQL 2005 presentation, while it
was Beta 1. Had funny talk with MS people (Russian
office) :-)

--
Dmitri Kouzmenko, www.ibanalyst.com