Thursday
Jan082009
Paper: Sharding with Oracle Database
Thursday, January 8, 2009 at 2:33PM The upshot of the paper is Oracle rules and MySQL sucks for sharding. Which is technically probable, if you don't throw in minor points like cost and ease of use. The points where they think Oracle wins: online schema changes, more robust replication, higher availability, better corruption handling, better use of large RAM and multiple cores, better and better tested partitioning features, better monitoring, and better gas mileage.
Reader Comments (8)
While I agree with you on cost, very probably even tco, there is no denying Oracle does a lot, if not all, of these things better/faster(/stronger).
They also have a valid point stating the features that MySQL does implement, like partitioning, are still immature compared to their technology.
But in their efforts to minimalise MySQL's achievements and benefits they pitch a very important sales line:
"Facebook had to make 'improvements' to MySQL replication"
and
"Google had to significantly change MySQL replication to make it work"
Although the tone of these statements is negative, I don't see this kind of flexibility in their product!
For $50k a server I would expect Oracle to be better at some things.
I've not even read the paper yet, but the statement that "Oracle rules and MySQL sucks for sharding" is totally incorrect and biased.
Sure MySql has some problems, however, many many large scale public websites use MySql with sharded databases.
I'd be interested to know of a large scale public website which can afford to use Oracle.
Ok, i've read it... but, MySql costs 0$ and oracle!?
In a web Startup this is very important!!! :-)
Hmmm....
Seems that all of those points are also features of Postgresql, except replication.
Imagine a shard of 1800 1 core oracle EE databases x $ 40K = $ 72M wasted only in oracle databses licenses,
OSE one does not have partitioning, replication, etc.... only this features exists in enterprise edition.
This configuration would lead to any company to bankruptcy.
This seems to be afraid of oracle to new approaches that left out their expensive database, the same apply for MS SQL Server.
Who is the audience of this paper?
There is no denying that Oracle enterprise-edition database is rock-solid, I use it daily and it consistently delivers. It's also $48k per processor (2 cores) at list price (expect a 25%-50% discount) for a perpetual license and a 18% annual support fee. Note that it went from $40k to $48k last August, wonderfully timed with a slowing economy.
So either you have deep pockets, in which case you're probably already an oracle customer and you don't mind spending a lot of money on licenses; or you're a small, growing company that needs to accommodate the addition of new servers often and would rather save cash and spend more time on operations and run mysql.
Unless Oracle significantly changes its licensing model the argument that Oracle is better for sharding is a very tough sell from a business standpoint. And within 5 years chances are, the RDBMS will be completely commoditized, just as the OS market ended up.
Although I've been a long time user of Oracle (since '94), my company at my insistence abandoned an Oracle Enterprise infrastructure in 2003 and migrated everything to PostgreSQL. There were certainly some obvious deficiencies in the beginning, but PostgreSQL has improved by leaps and bounds. We save hundreds of thousands per year even on what you'd consider a relatively small installation - the licensing and support fees were certainly a big chunk, but don't forget having enough Oracle DBAs to cover a few rotations at $100k++ each.
There are still some features in Oracle I'd love to have, but when I look at the overall price tag, I question my own sanity for even thinking about it.
All I'd really love to see in PostgreSQL is better replication - the add-ons are less than optimal at this point.