george at alink.co.za
Fri May 9 02:22:09 BST 2008
On Fri, May 9, 2008 02:27, Peter Hickman wrote:
> Steve's issue was with a complex query joining 8 tables (or
> thereabouts). Had you said that you had done something similar on MySQL 5
> then I would take it as an informed opinion on the subject. Sure 5 is an
> improvement on the previous versions but that does not mean that it
> addresses Steve's issue sufficiently.
I've done joins on many more than 8 tables and kept the query fast. This
of course is completely irrelevant since I'm not using the dataset than
> As for Google. Are they doing complex queries joining 8 tables and what
> sort of performance are they getting or are we just assuming that anything
> they touch is blessed?
ISTR Google use a custom version of MySQL 4.1 with added support for
synchronous replication, synchronised binlog positions and master/slave
promotion capabilities. Based on presentations from the last MySQL UC, it
would seem they're currently working out how to scale InnoDB to 16-32
cores and a 128G buffer pool.
As expressed in other emails, the main reason the whole MySQL vs PG debate
tires me is because MySQL users have nothing better than "PG doesn't
replicate" and PG users have nothing better than "MySQL has no
Both sets of users also use the classic "My query runs slow on (MySQL|PG)
but fast on (MySQL|PG)". This can usually be interpreted as "I dont know
how to make my query fast on (MySQL|PG)".
There's currently a hoohah on Google groups about people saying the
BigTable datastore is slow. Of course, this is because they're treating
it like an rdms, which it is not.
My current boss uses Coldfusion for everything, and I mean *everything*.
.. usual comment about only having a hammer.
More information about the london.pm