Re: To San or not to San

> Small database, less than about 150Gig of data (and that's 12 years of stuff!)

You mention you have 12 years worth of data. What was the data set growth per year? For the sake of explication, let's take the very simple (and probably inacurate) assumption that you have 150 GB divided by 12 years which gives us 12.5 GB of data per year.

Now, if you purchase a machine with six internal high speed disks of 1 TB and mirror three sets of drives then stripe all of them, that would give you about 3 TB of raw disk space. That's 3072 GB of storage. Let's keep this simple again and set aside this 72 GB to the OS and the data « lost » when we actually build the RAID configuration and filesystem. That would give us 3000 GB of space. Minus the 150 GB already in production, we're down to 2850 GB of disk space.

Unless a new project comes along which suddenly starts generating tons of data, if you start filling this 2850 GB of disk space at the rate of 12.5 GB per year, it would take 228 years before you get a file system full error.

Now my math is very, very basic. But even if we make it accurate, it would still take a lot more time to fill up this new volume than the normal service time for a server. Which is about more or less five years.

Next thing to check is the price. Which of these solutions is cheaper :

1. upgrade the current SAN with a new disk shelf? (can it be done?)
2. purchase a new SAN? Have it installed and configured. (I doubt that, but you never know what kind of discount a local government gets :)
3. purchase a new machine with six internal disk drives and a decent hardware RAID controller.

Don't forget to calculate the cost of the electric bill for these scenarios and the load on the HVAC system. The real estate in your data center (you have enough space in a rack?). The ethernet switch port count (do you have enough ports to have the new setup online at the same time as the current one?) Same ports question for the SAN fabric switches.

> So of course, now they are running out of space. They're starting to throw wild ideas of getting another SAN just for Oracle, etc.

If you do have a lot of money to spend, a SAN just for Oracle could be a nice toy. But IMHO it would be quite expansive.

> 2) Is S.A.M.E still the right philosophy?

IMHO I think so. Oracle ASM and RDBMS documentation also think so.

> 3) Is S.A.M.E really relevant or needed given todays high-Speed SANs?

It does make a difference. I've been testing various LUN configurations with Oracle RDBMS 11.2.0.3 with volumes on an HP EVA 4400 SAN and I/O patterns are quite different depending on which RAID set the LUN is built.

HTH, DA+

--
David Robillard
Senior Systems Administrator
CISSP, RHCE, SCSECA, SCSA
http://www.linkedin.com/in/davidrobillard
http://itdavid.blogspot.ca/
--
http://www.freelists.org/webpage/oracle-l