In the beginning was The Mainframe. It came from IBM, as did all the peripherals… at an outrageous price. Along came an upstart called EMC who could sell us disk to work with The Mainframe. Not only did the EMC disk cost less than IBM’s, it could also be connected to our Unix systems. And thusly, did the EMC arrays come forth and multiply, until they numbered 13.

Because of the cabling restrictions of the day (20M or 40M maximum SCSI cable length, depending on server and storage), servers were physically located close to the storage, and storage arrays were scattered across the data center. If one array ran out of capacity it was unlikely that another array with free space would be close enough to be usable by the servers, if the servers even had space for another storage interface card. Toward the end we were using FC-AL (direct Fibre Channel) links between storage and servers to get around distance issues.

Eventually, a “simpler” solution emerged – a Storage Area Network. While many advantages were touted, the main attractions for ITSS were:

• Storage and servers could be located anywhere in the data center
• Servers could connect to multiple arrays via a single link (dual links for redundancy)
• Adding new arrays to the network was easy
• The new links were faster

(Full disclosure: I was one of the advocates for SAN within ITSS)

NAS was not considered at the time since most of our systems were using local filesystems, the data center network was only 100Mb/s, and gigabit ethernet was at least as expensive as fibre channel, and provided less performance since there were no offload engines available. NFS insecurity was also a huge problem – the data center network was flat, and very few hosts had more than a single NIC. Lastly, the migration from direct attached disk to SAN was more straightforward than a migration to NAS would have been, although it did take a year to fully migrate all the data.