external SATA disk array VS external SCSI disk array

I am setting up a two server cluster that will use an external storage array as shared direct attached storage.  So there'll be two application servers sharing an external storage box.  I want to purchase from one of the larger vendors (HP, IBM, Dell) so that I can get a high level of service on the hardware (with same day turnaround on failures).  My data center will be in Chennai, India.

I'm currently looking at an IBM solution which would use the EXP 400 which is a direct attached shared SCSI storage disk array.  It can take up to 14 drives.  As i understand it - the largest SCSI drive is 150 GB - so this limits me to around 2 TB.  Now we're goign to need a lot more than that over time.

What i want to know is:

a. is a direct attached SATA storage array an option?  (with 250 GB per drive - that would at least increase the ceiling)  Are there direct attached SATA arrays available from the mainstream vendors?  as i said - i need the high SLA.

b. can a server have multiple SCSI storage arrays attached to it?  i.e. if my application servers had two raid controllers, could they be attached to two external SCSI arrays?  thereby doubling the storage available to my cluster?

c. would SATA be a bad choice because of lower througput than SCSI and lower disk speeds than SCSI?

Who is Participating?
trever_macphersonConnect With a Mentor Commented:
In short the answer is:

A.  SATA storage is an option.  I have not seen this solution yet as a DAS unless it has Fiber Channel on the backend, that is the servers would require an HBA (Host Bus Adapter) installed.  There are very inexpensive Fiber Channel based SATA solultions out such as the AX100 from Dell and EMC.

B.  You can typically expand most SCSI arrays by adding a second (or more) cabinet to it.  Some solutions will grow up to 7 cabinets which can give you up to 8TB of possible growth.  

C.  This depends on what your cluster is running.  Is this an application cluster or a high-availability cluster.  If this is an application (Exchange, Oracle, SQL) then the number of I/O's may actually be too many for SATA to handle over time.  If this is a failover - HA type cluster for File and Print, then SATA should meet your needs.  

Another thing to note here is that SCSI tends to have a much better MTBF than SATA drives.  Which means that SATA drives will fail with more regularity than SCSI.  They are improving, but are not up to the pounding that SCSI can handle.  SCSI tends to have better read/write seek times than SATA and have larger on-board cache on the disks themselves to handle multiple requests.  Also important is that if you are looking at heavy transactions, then disk contention may become a problem if you go with larger but fewer hard disks.
kenshawAuthor Commented:
thanks for that!

in relation to B - where can i read more about this?  would i be able to to this with the IBM EXP400 that i'm planning on going with?  Is there special hardware to expand out to multiple SCSI disk arrays?  from what you've said - SCSI is probably what we're after.

We setting up a high availability cluster running a .NET app which also relies on SQL Server 2000.  The two servers will run in an active/passive failover mode using the shared storage.  The app is a storage application - so being able to scale up to 4 or 5 TB would be excellent.

Can you give me pointers on how I could expand our EXP400 out to multiple cabinets?

For the Dell PV220 (same as IBMs I believe) go to http://tinyurl.com/6ynjr, also check out http://support.dell.com/support/edocs/stor-sys/spv22xs/en/index.htm (you need to register)
For the IBM EXP400 go to http://tinyurl.com/7yhro 

As far as adding cabinets to the PV220 (the one I have experience with) it is fairly simple to do, a search at Dell's support site should give you details on how to do this, you have to register to get in.  Essentially you will need to have a RAID controller (any of the ones that support clustered servers - no all of them do) for external RAID will be needed in the server along with of course the cabinet with the disks.  It is a fairly basic and time-tested operation.  Not sure about the IBM, but I know with Dell you should specify when ordering that you plan on using it for a cluster to ensure that you have it sent to you ready for clustering.

Good luck!

Oh and the Dell scales to 8 cabinets and a total of 16TB, I imagine this is true across the board since SCSI is SCSI... they should all be capable of U320 as well.  If you are going higher transaction... go with smaller disks and more of them (such as 13 x 73GB drives per cabinet.. that will give you roughly 850GB of usable space per cabinet (assuming RAID 5).  Make sure that you have your RAID done right too... RAID choices are:
RAID 0 - Great for speed... no redundany or data protection, not really a RAID at all
RAID 1 - Terrific for reads, a bit slower on writes, but offers great protection as it is a complete duplication of the disk, usually good for cluster quorums
RAID 3 - Not usually used unless it is for streaming media.  This option uses the last drive in the set as the parity drive.
RAID 5- The most common, very economical, it uses a disk for parity information, however it is across all the drives in the RAID set so it provides you good protection.  Decent Read/Write speeds.
RAID 1/0 - Close to ultimate.  Very common in high transactional databases.. their are multiples of this, but typically it is RAID 0 set mirrored as a RAID 1 so requires a minimum of 4 disks (1 to 4 ratio) so it is not economical, but provides bullet proof protection combined with very fast R/W times.  Not common however for many common external SCSI DAS solutions.  Usually seen in SAN implementations.  The Dell PV220 does offer RAID 1/0 if you need it.

It sounds like you will want a small RAID 1 set for your quorum, and a RAID 5 for your data... you will not want to expand your RAID sets beyond your cabinets as this can have performance implications.  With RAID 5, if you can make your RAID sets 5 disks or 9 disks you fit well with the algorithm to get the most out of the solution performance-wise.
Question has a verified solution.

Are you are experiencing a similar issue? Get a personalized answer when you ask a related question.

Have a better answer? Share it in a comment.

All Courses

From novice to tech pro — start learning today.