Thinking of implementing a SAN? Not sure what a SAN is? How does it work with
SQL? I won't say I have all the answers, but I do have some recent experience
that you might find interesting. Keep in mind I don't claim to be a SAN expert!
References to storage will be primarily Dell hardware because that is what I use
- there are plenty of similar options by other vendors and I don't have the
experience to say one is better than the other, we just happen to use Dell as
our main vendor.
Probably a good place to start is with a brief (and simplified) overview of
the types of storage:
- Internal. One or more drives that actually are contained within the
server. Usually this limits your drive count to five or six due to space
constraints. Cheapest solution. Redundancy via RAID.
- External - Simple. Consists of a container that holds one to fifteen (more
or less) drives. Attaches to the server using SCSI, appears to the server as
internal storage. A good example is the Dell
Powervault 220S. Cost is reasonable. Redundancy via RAID.
- External - Network Attached Storage (NAS). NAS is really a dumb server
with storage built in. You plug it in, put it on the network, you've got
more storage. Appears to the server as a network resource (you'd map a drive
to it to use it). Another Dell example.
SQL can use this only if you enable trace flag 1807. Cost is reasonable.
Redundancy via RAID.
- External - Storage Area Network (SAN). It's not quite a server, but it
does have a built in OS. Storage appears as internal storage to the server,
but the SAN can support multiple servers. Highly redundant, RAID plus
redundant hardware (power supplies, fans, etc). Cost is high. Example is the
CX400 (really an EMC product). Greater complexity to install and manage.
Couple links giving you more details:
http://www.dell.com/us/en/esg/topics/products_nas_pvaul_nas_san_compare.htm
http://www.infrastor.com/tech/whitepapers.htm
http://www.mcpmag.com/Features/article.asp?EditorialsID=334
http://www.storageadmin.com/Articles/Index.cfm?ArticleID=8693&pg=2
Pricing depends on who you buy from of course, and your pricing tier.
Acquiring a terabyte of storage might cost you $20-25,000 at the low end for
external or NAS, six figures for a SAN. In very broad terms, if you need to add
space to your file server, NAS or just plain attached SCSI will work fine. If
you need performance and high redundancy, then think about a SAN.
The idea of a SAN is to give you a "pool" of storage. Need more
space, just add more drives. Not quite that simple, since drives need an
enclosure (a box to hold to them, provide power, cooling, connectivity) that
holds 10-15 drives. If the enclosure is full, you add another enclosure plus
drives, but to the SAN, it just looks like more storage. To give you an example
from my world, we went with the Dell CX400, starting with two enclosures, each
holding 15 drives. It's expandable to four enclosures, and beyond that it's
possible to do an in place upgrade to the higher capacity CX600. For the CX400,
the max logical unit (LUN) is 16 drives, so if I have a LUN configured with 5
drives and I need more space (or just more spindles), I can add drives to the
LUN, then depending on how the LUN is configured, I can expand the RAID set to
stripe across all the drives.
Redundancy is a big deal in the SAN world. Just about every component is
redundant and hot swappable. Done correctly, this means not just the SAN itself,
but the switches used to connect to it and the host bus adapters on the server (PCI
cards that handle the connection to the SAN). Cluster a couple servers, you have
cables going every which way! Performance is also a big deal. Newer SAN's will
use FibreChannel2 and have on board cache. With a large cache drive performance
isn't as critical (in theory!) but you should still buy disks with the highest
capacity and speed you can afford, and keep in mind that spindles still
matter.
SAN's also offer some other capabilities you don't usually see with other
storage types. Different vendors may use different names, but the three main
types are snapshot, clone, and mirror. Snapshots are a virtual copy of a drive
that is completed in seconds. Clones are a true physical copy and the time it
takes is based on the amount of data and the drive speed. Mirroring is
synchronous replication at the disk level, typically done with a remote site as
far as 60km away.
Configuring the SAN initially isn't simple. Ideally for even the simplest
install you need two HBA's installed per server, two switches (Brocade
and McData are common vendors - expect to
pay around $1000 per port), additional software to handle failover if any link
fails, switch zoning, plus setting up LUN's, setting up RAID, and don't forget
just the time it takes to get it all into a rack. I was lucky enough to attend a
four day class on the CX400 - it's not rocket science, but I don't think I would
want to try to do it from just the book either. Once we returned home we still
had a Dell/EMC tech do the actual install while we watched - worth every penny.
Once we got the SAN running, everything works the same. SQL sees the storage,
it just works. The SAN itself should require very little admin once it's running
(or so we hope).
So if you're thinking about buying more storage and you've decided a SAN is
the right way to go, I've got a couple suggestions that might help:
- Once you pick a vendor, go to the class before you purchase if you can. We
didn't have a choice on vendor (I'm not complaining, Dell has been good to
work with), but I hated buying something when I didn't fully understand it's
capabilities and limitations. Definitely your network dude needs to go, try
to get the DBA in as well. You don't need to master it all, but you do need
to know enough to make recommendations about what goes where. Classes aren't
cheap, neither is the SAN. Don't be penny wise.
- Cost is always an issue. We had a choice of 36g/10k, 36g/15k, 73g/10k
drives. I would have preferred the 73g drives but the price was prohibitive
if I still wanted 30 drives. Went with 36g/15k. Maybe I could have gotten
away with 15 of the 73g drives, but fewer spindles didn't feel right - a
performance decrease wasn't something I could risk.
- Plan for the space, power, and cooling you'll need. Even if you have
sufficient cooling, you may need to rearrange equipment or add a fan to
eliminate hot spots. Right now we're seeing a 15 degree difference from one
end of our server room to the other. Put two racks containing six servers
(two of them 8 proc boxes) plus 30 drives, you generate a LOT of heat. If
you don't have a standalone temperate warning system, get one. You lose AC,
the temperature is going to climb in a hurry. They should shutdown if
they get too hot, who wants to risk it?
I'll be glad to answer questions about the Dell equipment or do a follow up
going into more detail if there is enough interest.