Sometimes the cloud cuts costs even if you don’t use it

I was having a discussion with one of our Windows administrators a few weeks ago about Exchange 2010, which makes some pretty substantial departures from how Exchange did things in the past. Since I’m mostly a Linux and VMware guy I don’t want to get too much into the product itself (I’m sure the MS Exchange Team has plenty of that), but the biggest change is that instead of four clustering modes, Exchange 2010 has one, and it doesn’t require any shared storage. The gist of it is that databases are organized into Database Availability Groups, and the databases that form them are replicated around the DAG in an arrangement rather like a big distributed RAID-10 array, with the difference that databases and not logical blocks are being striped across the replication group. When a server goes down, each database pops back up on another node in the cluster. Because of some back-end database improvements (namely, the elimination of single-instance storage, which stores only a single copy of a message or attachment sent to multiple users in the same database), Exchange 2010 cuts down random disk I/O by a huge amount, making it much simpler to run on commodity direct-attached disk with little to no penalty. Combine this with the removal of the shared storage requirement, and you no longer need a SAN to run clustered Exchange.

(Before any Exchange people chime in: yes, I’m aware that continuous copy replication/log shipping has been available since Exchange 2007. It just wasn’t viable for larger environments because you couldn’t easily distribute where the databases got replicated, meaning you either ran a replication slave for each Exchange server or seriously overspecified/overcommitted your hardware.)

Microsoft minces words pretty frequently to save face with customers (as most corporations do), and they’re still pushing the opinion that high-end SAN storage is a good idea so as not to rock the boat and upset anyone who already shelled out for high-end SAN hardware to run Exchange. However, the truth of it as far as I can surmise is that Microsoft specifically redesigned Exchange to work on commodity hardware in order to cut operating expenses on their own hosted Exchange offering.

Many applications are seeing a major paradigm shift towards distributed processing like Hadoop and schemaless NoSQL distributed data stores like,  MongoDB and HBase, and proprietary software vendors are starting to take notice and move towards better use of commodity hardware. When there’s a lot of engineering effort involved, though, sometimes the best incentive for a company to improve the efficiency of their products is to try to make money on it themselves, and the results can benefit everybody.