EU clears Oracle's proposed acquisition of Sun Microsystems [LWN.net]

EU clears Oracle's proposed acquisition of Sun Microsystems

Posted Jan 22, 2010 16:41 UTC (Fri) by Simetrical (guest, #53439)
Parent article: EU clears Oracle's proposed acquisition of Sun Microsystems

Any serious user of MySQL is already using the InnoDB storage engine. (The
default storage engine, MyISAM, is non-transactional, supports only table-
level locks, and is overall totally unsuitable for most serious load
patterns.) InnoDB is GPL+proprietary software owned by Innobase -- which
was acquired by Oracle in 2005.

Oracle could already have crippled MySQL by killing development of its only
serious storage engine. It didn't. It funded big improvements of InnoDB,
including major performance and scalability improvements. There's no
reason to think it will turn around and do something different to MySQL
itself. To the contrary, maybe it will now make InnoDB the default storage
engine as it deserves; MyISAM only really makes sense for special-purpose
loads, and it's probably been the default because MySQL didn't own InnoDB.

Oracle is a company whose goal is to make money. It currently only has a
super-high-end database product. MySQL expands its range of offerings by
giving it the leading low-end product too. If it encourages MySQL to grow,
it will own both the top and bottom of the market, giving it more leverage
to squeeze out competitors like MSSQL in the middle. If it kills MySQL, it
loses a valuable asset; the low end will go to a fork or PostgreSQL. It
has no rational reason to do this. MySQL doesn't seriously compete with
Oracle. Not to mention Oracle's made legally binding commitments that it
won't kill MySQL, at least not soon.

I'm much more interested in what will happen to Solaris. Oracle has been a
prime backer of btrfs, for instance, whose only real goal is to give Linux
a ZFS-like filesystem. Now Oracle will own ZFS. Will it see any reason to
keep its assets incompatibly licensed, or will it GPL Solaris and let ZFS
(et al.) get ported to Linux? btrfs won't be as stable and featureful as
ZFS for several years yet.

(Log in to post comments)

EU clears Oracle's proposed acquisition of Sun Microsystems

Posted Jan 23, 2010 4:14 UTC (Sat) by lakeland (subscriber, #1157) [Link]

I disagree on a few points.

Firstly, there are still some serious uses of MyISAM - it's very fast for dumb things, so when you
don't need transactions, locking, etc. then it provides a good option. Also, it wasn't that long
ago that some of the fancier options (e.g. replication) were only supported by MyISAM. I do
agree that InnoDB is better in most instances.

Secondly, if they're going to change the default storage engine then I hope they go with Falcon
rather than InnoDB. In my opinion it is substantially better.

Thirdly, Oracle has a number of database products (Times 10 and Berkeley) and is not remotely
averse to killing off products when it sees a long term advantage in doing so. Just look at how
terribly they deal with products like Discoverer, Forms or Designer.

I don't know what will happen with ZFS - When I tested it in a deliberate hardware failure I
managed to corrupt all of the data after removing just one drive, which put me off exploring it
further (that was almost five years ago). I'm really holding out for a proper distributed filesystem
and Sun has much more potential to release one than any other company. It's possible that could
fit well with RAC or else Oracle may see that as competition as it brings replication to a lower
price tag - hard to know but that's what I'm going to be watching for :)

EU clears Oracle's proposed acquisition of Sun Microsystems

Posted Jan 23, 2010 23:50 UTC (Sat) by Simetrical (guest, #53439) [Link]

1) I did hedge by saying "most serious load patterns". Yes, if you're
doing inserts and selects only, MyISAM is fine. As soon as you have a
significant number of deletes or updates, its performance drops like a rock
due to lock contention -- any pending update or delete will block *all*
queries until a) all currently executing select queries have finished (so
the update or delete can begin), and then b) the update or delete finishes.

MyISAM has other critical flaws. For instance, if mysqld crashes, or even
sometimes if queries get killed, tables can become corrupted. This
requires a repair, which can take multiple hours on a large dataset.

Replication is (usually) statement-based in MySQL, so as far as I know,
it's always worked for any storage engine. It's certainly worked on InnoDB
since at least 4.0 (released in 2003), and I know that for a fact because
Wikipedia used 4.0, InnoDB, and replication until very recently.

MyISAM has a few advantages over InnoDB. For instance, it supports
fulltext search -- but for real work you don't want to use it, you want to
use an external search provider like Sphinx or Lucene, since those perform
vastly better. It takes up less disk space, but that's usually only an
issue with stingy shared hosts that limit your database size; disk space is
cheap. There are a few other goodies like prefix compression of indexes
that InnoDB doesn't have yet.

It makes sense to use MyISAM in some limited circumstances. But without
InnoDB, MySQL *would* be crippled, and it *would* make more sense for
InnoDB to be the default storage engine so MyISAM is relegated to the niche
use cases it's good at. So my point holds.

2) Falcon is dead, isn't it? The creator of Falcon, Jim Starkey, quit
MySQL in June 2008. It was meant to be the hallmark feature of 6.0, but
now the MySQL website officially says 6.0 is no longer being worked on.
Resources have shifted to Maria and InnoDB.

In any event, it seems premature to compare Falcon and InnoDB when InnoDB
has been in production use for a decade or so, while Falcon never
progressed beyond alpha. You can't call alphas "substantially better" than
shipping software; they aren't even usable.

3) Sure -- but they haven't killed InnoDB, and could have done so at any
time to MySQL's massive detriment. It's unlikely they'll try to kill
MySQL. It's possible, but then it will just be forked, or pgsql will take
over, or whatever. I don't think it's much to worry about.

4) Yeah, ZFS still has lingering reliability problems. But it's still
several years ahead of btrfs. Even if it's not quite ready for some sites,
it will be ready a lot sooner than any comparable alternative (and btrfs is
the only one of those).