After the arrival of Exchange 2010 and its DAG feature many people suggested – Microsoft included – the option to run Exchange on low-cost SATA disks in JBOD (Just Bunch Of Disks) configuration, provided you have at least 3 database copies. As you probably know, DAGs enable you to have multiple copies of mailbox databases running on multiple servers with a configurable lag per copy.
This suggestion to use JBOD, as well as the discussion of going backupless or not, isn’t without controversy. For many years people have learned to put their (critical) data on redundant storage. With Exchange 2010 this dogma is said to have changed, because contrary to its predecessors, Exchange 2010 can happily run fully supported on low-cost SATA storage in JBOD configuration. The argument used is that because you have at least 3 copies on 3 different physical servers you can survive a single failed mailbox server (likely) but also two failed mailbox servers (unlikely?).
The first problem is underestimating the limits of relying on 3+ copies. Nobody expects the Spanish inquisition 3 failures, right? But what if your hardware (e.g. disks) are from a faulty batch or contain buggy firmware, and you equipped all three of your physical servers with those parts (something burn-in tests should bring to light, but who does that these days?). I know of several occassions where drives died in pairs within hours from each other; you better make haste recovering your mailbox then. Is this perhaps a reason to look at different vendors, different parts? After all, this is more or less the same reason many businesses require multi-vendor products when reliability & security is concerned, e.g. anti-virus products from different vendors. The idea is to spread the risk.
Also, being able to use JBOD (and go backupless) looks interesting on paper, but don’t forget that – as suggested – you need to get yourself at least 3 physical servers (and no, don’t run them virtually on the same host). So, in the end this may lead to less servers (with RAID) being the most cost-effective alternative when looking at the total picture, e.g. hardware, licenses (OS, Exchange, AS/AV agents, management software, etc.) and operational costs. Why run (and maintain) many servers when the additional costs don’t outweigh the benefits?
A third element of the JBOD versus RAID discussion is the time to recover the original situation. When one of the servers fails, you should rebuild the server (hope you have some spare parts lying around or a decent replacement service contract). And after rebuilding (or restoring) the server, you need to reseed the database copies. This step may take a long time, depending on the size of the databases. Replacing a harddisk and rebuilding RAID sets is much quicker and much easier (and less prone to error).
In the end – as always – the choice should be based on business requirements. Perhaps your business can do a few hours without e-mail while IT is recovering services (can’t imagine, but you never know). In that case it’s nice to have a supported low-cost JBOD/SATA option. In my opinion, the benefits of having proper RAID setup outweighs the trouble you have to go through when repairing your JBOD based solution. Depending on these requirements, and how deep your pockets are, I’d go for a combination of RAID and DAG, where RAID is used for availability and DAG for availability (same data center) or resilience (multi data center, i.e. disaster recovery).
Oh and one other thing: when you must, use proper “Enterprise class” SATA disks; they’re made to run 24×7.