Re: Database market share 2004

Jurgen Haan wrote: > DA Morgan wrote: > > Jurgen Haan wrote: > > > There is another important reason too: Instrumentation. If they are > > slow diagnosing why is a question of making guesses. That may be a > > reasonable approach when supporting a small non-commercial web site. > > It is a non-starter when talking terabytes and a requirement for 7x24. > > And then there's that little problem with government requirements > > around audits. They just aren't ready for prime-time. > > 24x7 is actually no problem with OSS db's like Postgres. > At the company where I work we have a DB2 DB and a Postgres DB running. > Neither of them have to be taken down during maintenance. Next week I get to swap out two storage arrays on a 500-gbyte db2 udb database server: the new arrays are larger and faster. Anyhow, one nice feature about db2 is that I can add the new arrays & remove the old ones in a relatively simple operation in which the database handles rebalancing all data automatically, and everything is online the entire time. Didn't think postgresql was up to that yet. > What's a big problem with postgres (and actually one of the main reasons > why we don't use it for our sensitive information) and that's that > postgres is extremely unrelyable in high TPS situations. > Scalability with OSS databases just plain sucks (if any). Although I really like postgesql, the biggest reasons not to use it in production for us are: 1. If it can't do everything then we need db2 as well. Standardizing on db2 is simpler than maintaining a mix of skills, procedures, products, etc. 2. postgresql doesn't support query parallelism or partitioning. This means that db2 table scans run about 40x the speed of postgresql table scans (assuming a 4-way server and typical partitioning benefits). 3. db2 is pretty inexpensive in its small-server editions (workgroup, express, etc). So the savings of going with postgresql is more than wiped out by the additional hardware requirements. > But still I think OSS databases are to be reconed with. > > It's the same as the early 90s, Linux, what a cute little project, but > it surely will never be of any importance. Now, just take a look a the > linux population among Internet web servers. Yep, postgresql is on track to be a cool DBMS. buck

Buck Nuggets wrote: > > Next week I get to swap out two storage arrays on a 500-gbyte db2 udb > database server: the new arrays are larger and faster. Anyhow, one > nice feature about db2 is that I can add the new arrays & remove the > old ones in a relatively simple operation in which the database handles > rebalancing all data automatically, and everything is online the entire > time. Didn't think postgresql was up to that yet. > Postgres doesn't have to be. The storage arrays are getting smarter aswell. Take the NetApp storage array, you can expand, reassign disks to a logical volume, resize a logical volume online. And I'm not sure if it can be done online, but the head (the controller part) should be hotswappable aswell. This way you can upgrade your 1TB netapp to a 6TB netapp. But if you're using a simpler array, a simple SAN or NAS, it's indeed very pleasant to have the ability available in your database. -R-