[Sunhelp] Software RAID or Hardware RAID...

[Stephen Holmes]

You're looking at a typical trade off here.

Software Solutions

Sun's disksuite is (I believe) free with the server version of the Operating
system (You're licensed if you bought a server).
There's also Volume Manager by Veritas, I have less experience, but Sun's
feedback has suggested that this is a bit more "Enterprise scale".

Experience shows that these are pretty good if you're keeping things simple.
I have however seen performance drop off under load.
Striping increases potential disk throughput, and you can make dramatic gains by
tuning interlace sizes to match your particular traffic mix.
Mirroring gives resilience (but at 100% hardware overhead).
RAID 5 (which is what a lot of folks mean when they say RAID) is not
recommended.
Remember that it's your CPU doing all the work to calculate parity, and you
really want those cycles handling Apps.

Hardware Solutions

More cost and some dedicated hardware (usually the type of compute power you'd
have had on you desktop before the last machine upgrade).
Your CPU is no longer loaded by calculating parity etc.
Hardware is really the preferred solution if you want to use RAID 5, and the
savings on using only 20% disk overhead may pay
off some of the costs of the rest of the hardware.
Some boxes perform RAID 5 only, others have a much smarter controller and allow
lots of fancy combinations.

Things to watch for.
With computation taking place away from your server, you have a new point of
failure in your system.
Good systems cost more, but have features such as
     Dual Power Supply.
     Redundant Data Patchs.
     Dual redundant controllers
Unless you have these, then you're only protected against failure of a single
internal disk.

On the subject of single disk failures, most decent RAID units incorporate a Hot
Spare facility.
If It implements 5 way RAID, then a 6th drive is stored in a live bay.
When a disk failure is detected, the system begins to synchronise the spare, so
there;s no downtime or manual intervention to initiate recovery. Later on you
hot swap the failed drive for another spare.
This is important with RAID 5 because the system can run with 4/5 drives active,
but performance is heavily degraded when running like this.

Remember to check upgradability of the RAID system.
5 disks giving 72 Gbytes sounds pretty cool now.
In a few years it will look pitiful, and with the higher level of investment,
beancounters will expect your "Storage Box" to depreciate a little slower than
usual.

The RAID unit can become a bottleneck in several ways, notably:
     Single connector limits total throughput, and there's a tendancy to pile
far more data into a RAID unit than you would a JaBoD chain.
     Processor speed doubles about once per year. Can the processor (and cache
memory) in your unit, be upgraded to keep pace.
     My experience is that on-board memory is critical to the performance of HW
RAID.


Steve Holmes