Disk in ? disk out ?

(I think I?ve watched the Karate Kid too many times)

System builders are often faced with the dilemma of adding enough drives to a system to meet a customer?s capacity and performance requirements. The traditional approach has been to purchase a chassis large enough to hold all the drives internally. This approach has a few problems.

Firstly, you need to spend up big on the initial chassis. Even if the customer doesn?t need all that space right now, if you are going internal then you still have to spend enough money up front to purchase the 24 or 48-drive chassis.

A smart customer will want to purchase a system that can grow with his needs, but purchasing a massive server up front often puts the quote out of the reach of the initial sale.

Purchasing a smaller chassis without expansion capability often means people are thinking of replacing the hard drives at a later date with larger drives to meet a capacity upgrade requirement. Bad move ? this is not cost effective, takes a lot of time and is not in the best interests of the customer (even though it may get the system builder a bit of extra income at some future point in time).

So ? you want to build a system now that meet?s today?s requirements for capacity, and you only want to pay for what you are using now ? smart. But ? you know your space requirements will grow in the future and you don?t want to throw away a perfectly good server just because it has run out of disk space a year after purchase.

What to do?

It?s pretty simple really ? just put the disks ?outside? the server. Many users can purchase all the horsepower and memory requirements they need in a cheap 1U server. By adding a JBOD (just a bunch of disks ? I love that acronym) and a RAID card with external connectivity, you can add as many drives as you like. On top of that you can just keep daisy-chaining JBODs together until the cows come home, giving you as much storage capacity as you can possibly imagine (try 250TB in the one server).

If this is so simple, why do people shy away from it? FUD (fear, uncertainty and doubt) is generally the reason. Let?s look at some of the questions people have when considering this type of configuration:

? I?ll create a performance bottleneck between the server and the storage ?
? If one cable breaks I?ll lose all my storage ?

A modern SAS RAID card (and yes, it accepts both SAS drives and SATA drives at the same time), has a funny looking new type of external connector called a 4x (4 by, 4 lane, 4 channel ? call it what you will) connector. This is, practically, 4 single SAS/SATA communication channels joined together at the hip.

A single SAS/SATA connector runs at 3 gigabit ? roughly 300mb per second throughput. But SAS is smart, much smarter than SATA. When a SAS card finds a 4x cable going from the card to another device that also uses a 4x connector, it joins all 4 cables into one large pipe. Sort of multiplexing for those of us old enough to remember that technology.

The end result of all these smarts, is that the pipe between the RAID card and the JBOD is capable of sustaining 1200mb per second throughput. That?s pretty fast, and eliminates any bottleneck concerns that a user may have about connecting drives externally to a card.

Now the question about the cable. When was the last time you saw an external cable ?break?. I?ve spent a lot of years in tech support, especially dealing with SCSI. Now SCSI connectors were dodgy to say the least, and were (and still are) often used on devices like external tape drives where they were connected and disconnected frequently.

Over the years I have hardly replaced a SCSI cable. Now if we add to that the fact that modern SAS external cabling uses a much better connector type and cables connecting servers and JBODs are not connected/disconnected frequently, and you end up with an almost bulletproof connection system that is by no means the weakest link in the data storage chain.

Now that we?ve dispelled the myths (FUD), let?s look at the benefits of building a server system in this manner.

Firstly, you start with a 1U server with 4 drives (or 2 or 3 and a CDROM). Spend your money on CPU and memory components. In that box you need a SAS RAID card that has 4 internal connectors to handle the (up to) 4 drives inside the box, and a 4x external connector to connect to a JBOD. An example of a card like this would be Adaptec?s 5445.

Then buy yourself a JBOD. These are generally 2U rack units that take 12 drives. You may or may not want to purchase all 12 drives now ? just purchase as many as you need. Therefore your initial investment in the server hardware comes down to a beefy 1U server, a good quality RAID card, a JBOD and as many drives as you currently require.

But time goes on ? and sooner or later you?ll run out of space. Simply purchase more drives and add them to the JBOD. Fill the JBOD? No worries, purchase another JBOD and daisy-chain it to the first one. Start filling that with drives. Run out of space? Just repeat the cycle with another JBOD and keep on adding drives for as long as you wish. The technical maximum at this point in time is 250 hard drives, which for all intents and purposes is unlimited ? I haven?t see anyone come even close to this yet.

So you end up with a big expensive server ? great. However you paid for it over a couple of years, and took advantage of the dealer specials along the way. You also gained from hard drive price decreases. Along the way were hard drive capacity increases (these two go hand in hand like death and taxes).

It?s hard to see anything but benefits to users ? reduced capital expenditure, increased flexibility over a fixed system, access to better technology as it both appears and drops in price.

Sounds like a win win win to me.