Weekly Edition
Return to the Press pageSponsored link
Vyatta –
Linux & Open Source
Alternative to Cisco –
Advanced Routing,
Firewall, VPN, QoS..
Free Download ->
| |||||||||||
Jim Starkey joined MySQL AB
Jim Starkey is the original creator of InterBase which became Firebird.
Here's blog entry at Firebird News in which he made it publicly known that
he now works for MySQL AB. "My company, Netfrastructure, Inc., has
been acquired by MySQL, AB. As part of the agreement, I will be working
full time for MySQL." (Thanks to marius popa)
(Log in to post comments)
Shopping for backends Posted Feb 20, 2006 20:58 UTC (Mon) by man_ls (subscriber, #15091) [Link] Hmmm, is MySQL AB already shopping for a new transactional backend for MySQL?
Shopping for backends Posted Feb 20, 2006 21:54 UTC (Mon) by tetromino (subscriber, #33846) [Link] Not shopping for. They are (I believe) writing a new one from scratch, so they can dual-license it themselves. After the InnoDB debacle, they certainly don't want to leave the backend under the control of a different company, and a purely open-source backend doesn't fit into their business plan.
Shopping for backends Posted Feb 20, 2006 23:25 UTC (Mon) by eyal (subscriber, #949) [Link] And with Jim Starkey then they are going to have a top engine. Firebird, and it's commercial parent Interbase that Starkey created, have examplary transactional capabilities.
EZ
Shopping for backends Posted Feb 21, 2006 7:10 UTC (Tue) by kleptog (subscriber, #1183) [Link] Indeed, judging by the description at here it looks like an MVCC model (PostgreSQL, Oracle, modern versions of MS-SQL, etc) which has a proven track record for good performence.
Shopping for backends Posted Feb 21, 2006 10:09 UTC (Tue) by drag (subscriber, #31333) [Link] I am not to familar with database stuff personally, but I've seen Jim Starkey refered to as a 'database god'. It looks like MySQL can become a top-line database server over the next few years if they play their cards right.
Some history from a quick google search..
Shopping for backends Posted Feb 21, 2006 11:46 UTC (Tue) by nix (subscriber, #2304) [Link] Oracle's MVCC is, well, horribly implemented. I don't know who had the idea of pushing old data off into a separate segment which had no usage bitmaps, but they were insane whoever they were. I mean, let's see. For the price of avoiding periodic VACUUM (a trivial ask in PostgreSQL 8.1 thanks to the autovacuum daemon), Oracle has to
- do a triple-write for the first UPDATE of a given row in each transaction (once to the log, once to the rollback segment, once to the datafile) as against PostgreSQL's two writes (once to the log, once to the datafile);
- have special code to determine that, hey, we should look in the rollback segment and not the datafile for this block (and in fact it incurs a read of the datafile to determine this, first, slowing reads in transactions whose rows have been updated by other transactions)
- add piles of administrative gubbins to allow applications to switch rollback segments to bigger ones if they think they're going to run out of space (which is nearly impossible to determine, see below).
The lack of any sort of usage-tracking information for rollback segments introduces really bizarre potential-data-corruption bugs; e.g. if you hold a transaction open for `long enough' (and you can't tell how long that is as it depends on levels of update activity in other transactions), the rollback segment will wrap around and *overwrite* the data the long-running transaction is using. Oracle spots this and kills it, but long-running transactions are of course exactly the ones you least want to kill! In PostgreSQL this is of course a total non-issue: the transaction is still in use so vacuum doesn't suck up its rows. (The analogue in PostgreSQL is running entirely out of disk space, but that requires that you have enough concurrent transactions to hold most to all of the space on that disk; the Oracle failure mode just requires that enough data has been written by *any* transactions, even long-dead ones, to wrap around the rollback segment.)
Oh, and it makes held cursors almost impossible to implement (I don't think Oracle has them yet).
|
|||||||||||