In 18 months we could see 2.5-inch drives supplant 3.5-inch drives in the performance tier in storage arrays. That's the view of IBM "master inventor" Barry Whyte at the company's Hursley House labs near Winchester. What's going on?
In quick succession we have seen Hitachi upgrade its USP-V top-end array to the VSP and introduce 2.5-inch drive support alongside 15,000 rpm 3.5-inch drives. Meanwhile, HP has OEMed this product. It calls it the P9500 and supports only 2.5-inch drives, abandoning 3.5-inch drives for the performance tier and bulk data storage. IBM recently introduced its Storwize V7000 midrange array and supports only 2.5-inch form-factor drives in that.
These are three straws in the wind, but their provenance is such that we should take note; the days of 3.5-inch drive dominance in the array performance storage tier are drawing to a close. It also suggests bulk data storage has no place in an enterprise drive array.
The end of 3.5-inch performance drive dominance was actually first signalled by Xiotech when it launched its storage brick -- the sealed ISE enclosure -- in 2008. Atrato also launched a 2.5-inch array product but has struggled to make headway since then, whereas Xiotech is progressing nicely. These vendors were in at the start and showed what could be done. But it will take the adoption of 2.5-inch drive technology by market-leading vendors such as Hitachi, HP and IBM for the trend to become mainstream.
We can now expect EMC and NetApp to follow suit in 2011, as well as Oracle, and that will tell the hard disk drive vendors that they should plough more resources into 2.5-inch drive production and development and less into 3.5-inch drive manufacturing and research.
Why use smaller drives? The reasons are threefold. Firstly, you can pack more drive spindles into a drive enclosure using 2.5-inch drives than you can with 3.5-inch ones. Drive I/O performance has been stuck for a decade or more, and yet the pace of business and server virtualisation means that servers want more I/Os from their storage. You can stripe data across drives to get more I/Os off a file or database, but that gives you a one-time boost. You can add solid-state drives for the really hot data, and that gives you another boost, but SSDs are expensive and you can't store all your active data in them.
That leaves an I/O gap, and the fix the industry sees is to move to more spindles so you get more read/write heads reading and writing data in a rack enclosure. Secondly, you can have more drives in a cabinet; the storage density goes up, the amount of floor space you need goes down, and your expensive data centre space is used more efficiently.
Thirdly, small-form-factor drives draw less electricity, and so there is an energy saving.
It's not all good news, though; 2.5-inch drives spin at 10,000 rpm, not the 15,000 rpm of the fastest 3.5-inch drives. That makes the use of an SSD tier of storage in 2.5-inch drive arrays good for really fast access to data. IBM's Whyte talked of short-stroked 2.5-inch drives perhaps exceeding 400 IOPS, but surely the days of short-stroking drive are coming to end as flash drives are affordable enough and so much faster.
Once 2.5-inch drives achieve around a terabyte of capacity, the move away from 3.5-inch drives for performance data storage should accelerate.
There are two approaches to the issue of limited capacity of 2.5-inch drives. One is to say that in the performance storage product space it doesn't matter; which is what HP and IBM are saying with their new products. It's half-hearted, though, as they aren't abandoning existing arrays that offer high-performance 3.5-inch drives, such as IBM's DS8700 and HP's XP24000. Their message with the new arrays is that you should put bulk data in externally attached arrays full of SATA drives.
This makes tiering data across performance and bulk data drive tiers trickier than it is inside a single drive array. There is also the issue that if you have bulk data arrays and performance data arrays in the data centre, the floor space and electricity saving you get from a 2.5-inch drive-based performance array is illusory. You still have one or more bulk data arrays, and these need power and cooling and floor space; and they have their own controllers, network links and adapters. You may well end up using more power and cooling and floor space.
The second approach is to mix and match 3.5-inch and 2.5-inch drives in the same array. This is what Hitachi is doing with its VSP, the technology base for HP's OEMed P9500. Hitachi can dynamically tier data across drive tiers inside the VSP and may well have a stronger product than HP's P9500 or IBM's V7000 when both raw performance and bulk data storage is needed.
Compellent told attendees at its user conference in May that it believed 900 GB 2.5-inch drives were coming. It also mentioned that it saw 15,000 rpm 2.5-inch drives coming too, with 300 GB capacity. Perhaps we'll see products like these next year.
Once 2.5-inch drives achieve around a terabyte of capacity, the move away from 3.5-inch drives for performance data storage should accelerate. The overall capacity of a 2.5-inch drive enterprise array should be sufficient for most applications and the array's overall IOPS count should blow a 3.5-inch-based drive away. At that point the disk drive vendors may well slow down -- and perhaps stop developments intended to add more capacity to fast 3.5-inch drives and let them wither away.
If 2.5-inch drives then head toward having terabyte platters, the hard disk drive vendors may investigate making four-platter models that could hold 4 TB and still have energy-draw and cooling advantages over 3.5-inch drives as well as being able to be packed more densely, although not as densely as two-platter drives. That would be a 2013/2014 story, if it happens at all.
Chris Mellor is storage editor with The Register.