Remember 5.25" hard drives? I remember 18" hard drives. Then 14", 9" and 8", 5.25" and now 3.5". The secular trend is clear: disk drives shrink. So when do we make the leap to 2.5" drives in the data center.
Traditionally, the jump started when the smaller form factor could hold a the operating system, applications and some user data. Obviously 2.5" drives passed that point several years ago and still no migration. Why?
Storage arrays are one reason. Arrays have made disks simple bit buckets instead of a design center for new systems. The shrinking number of I/Os per GB is hidden by cached controllers. Density per square foot is increased by drive capacity growth. And you may recall that the only 3.5" form factor available today is what used to be called Low-Profile, which is thinner than Half-Height. Thinner drives mean greater density.
Five out of five analysts agree: power is a problem On an analyst panel at Cisco yesterday, we got into power issues for today's data centers. How to fix? Power distribution, cooler CPUs, containerized data centers all got some play. To me though, all that is nibbling around the edges of the problem. Data capacity is exploding, so what about disk drives?
The new quad-core Xeon CPU draws less than 50 watts. Stick it in a server with 12 drives and the storage power draw drowns it.
Optimized for low power consumption Vendor power stats are about as useful as vendor MTBFs, given all the unspecified assumptions that go into them. Yet the peak numbers are suggestive. Enterprise drives, FC, SCSI and 10k SATA drives, are in the 12-15 watt area. 2.5" notebook drives are 2-3 watts. And don't forget that for every watt consumed, another 0.4 watts is needed for cooling, not to mention power supply and distribution losses.
GB per watt? Lower in 2.5" because the drives are slower and enterprise drives have similar capacities to support their higher speeds.
Cost per GB crossover Over at the market-driven prediction site Storage Markets they've been looking at the 10k 2.5" enterprise drive market. Last I looked they predicted that the cost per GB of high-end 2.5" drives would equal 3.5" drives some time this year. But Seagate's 10k 2.5" drives are power hogs, spec'd at about 8 watts peak. They'll give you a tiny and speedy array, but not a particularly power-efficient one.
The race is not always to the swift Array vendors like fast drives because they carry higher margins, but drive speed is just one way to achieve array performance, and not a very good one at that. Caching and striping are much more important, while big commodity storage systems - think Google and Amazon - use file replication. Even though their heads are slower, 2.5" drives pack many more heads into a given cubic volume, giving array engineers more degrees of freedom in crafting solutions.
Nor riches to the wise I have no doubt that engineers at Google, Amazon, Yahoo and MSN are calculating when it will make sense to move to 2.5" drives. Once they start the economics will complete the job in less than five years. Average data center floor space may even start to shrink, given the much greater space efficiency of 2.5" drives.
But time and chance happen to all men The economics of high-volume manufacturing are such that once the switch begins, the advantages will be almost irresistible. What triggers the switch is less clear today than in the past. There can be no doubt that power concerns will play a much bigger role than in past form factor migrations.
Comments welcome, of course.