In our case it comes down to tiers of storage service. This has come about in large part because different needs have different storage requirements. Our ESX environment has Exchange running inside of it, so we need fast, reliable storage. Our desktop-support function just needs lots of it (disk images), with no requirement for speed. The second type doesn't need the stuff that's $9/GB.
Tier 3: Homebrew NAS
This is an HP DL360 with four attached MSA60's and to-be-determined storage software. The drives are all 7.2K RPM MDL SAS drives, giving about 30TB of it. The software will be picked soon but is likely to be a combination of openFiler for iSCSI services with a Windows server attached (via iSCSI) providing file-level serving. Total cost-per-GB is in the neighborhood of $2.
Tier 2: EVA4400 - FATA
This is an EVA4400 with .5TB fibre-attached-ATA (FATA) drives, a 7.2K RPM highly reliable solution. This is only accessible via Fibre Channel, though iSCSI is an option. This is used for highly available file-sharing (by way of a cluster), mass storage of other types, and backup-to-disk stuff. Total cost per GB is in the neighborhood of $9.
Tier 1: EVA4400 - FC Disks
This is another set of shelves on the EVA that are running 450GB 15K RPM FC drives. This is used for storage that NEEDS low latency, high volume traffic, and can handle highly random I/O efficiently. The tenants here are the ESX datastores, our MSSQL database volumes, and certain heavily accessed file-serving volumes. Total cost per GB here is hard to pin down, but between $12-$17/GB.
The first tier is the newest one and it was a hard add. The whole point of it was to provide a centrally managed cheap-ass storage solution so individual departments wouldn't have to buy servers to get the storage they wanted. The hardware is all covered by warranty, but the software that drives it? Only in one use-case, and that use-case is not the one I recommended to management. We could have slapped a server running OpenFiler onto the EVA4400-FATA and served things up that way, but that still wouldn't have been cheap enough, we had to build ours from disk-array parts.
We have tiers of service for a variety of reasons, one of which is cost. The other is performance and expected I/O loads. The MSA60 based solution should saturate I/O wise a lot faster than either of the EVA options, simply because it has fewer spindles to spread the I/O around (vs FATA) or uses slower disks (vs FC). My testing on the MSA60-based solution shows that for some workloads (sequential) I'm hitting the SAS transfer limit, which is slower than our FC capable arrays are able to pitch data.