* Storage blades
Last time, we mentioned some of the values blade servers deliver – limited impact on floor space, device consolidation and centralized management. But the overwhelming majority of server blades are processor blades. What about blades for storage?
Several companies are looking to build storage blades into their blade server lines. Last March IBM committed to developing storage blades for its eServer line, and has already come out with an early version of a blade holding six 2.5-inch SCSI drives. According to IBM this blade is aimed mostly at providing boot-up services for the various blades within the chassis, and is not intended to be a general-purpose storage blade. But six devices could offer in excess of 600G bytes of storage; even mirrored, this seems like quite a bit of capacity for booting purposes.
HP is developing a Serial Attached SCSI (SAS) storage blade for its BladeSystem servers as a way to offer customers a scalable system for computing, networking and storage. Right now it looks as if the system will be ready for launch by year-end.
The choice of SAS for blade storage is an interesting one. SAS devices will be relatively cheap compared to Fibre Channel because they are well along in the process of becoming the first choice for direct-attached storage, and thus the per-unit costs should be starting to drop in the not-too-distant future. Additionally, SAS connectivity can help address some of the limitations that are sure to appear in a server chassis.
What? Limitations in a blade server? You bet.
For one thing, it is not at all clear how scalable the present generation of blade servers is going to be. Scalability will mean much more than finding room in the server frame for another blade; it will also require adequate amounts of power, cooling and networking capability.
SAS drives will help with the cooling issue in two ways. First, the smaller connectors needed for the serial link to the backplane will make some contribution to improving air flow (although the smaller form factor drives will make a much greater contribution). More importantly, the new generation of drives draw less power than the previous generation – a feature of the drive electronics and not a SAS feature, but nonetheless a feature that SAS takes advantage of. Scaling will thus be less stressful on power requirements and will run cooler as well.
The networking challenge will be no different than the networking challenge that comes with any large mass of storage – if the connectivity does not scale in proportion to the data that needs to be moved, throughput will be throttled and will never scale. The maximum amount of storage will thus be effectively capped by the total amount of data than can be moved on and off the server. SAS won’t help that, but there may be other solutions in the works.
Still, SAS devices would seem to fit pretty well with the needs of a storage blade. Chip builder PMC-Sierra thinks so, at any rate. The company has announced a full family of SAS “zoning switch” silicon, which will enable segregation between broadcast and SAS traffic, and which can ensure access control from any device group to any other device group – and that will lead to more easily managed resource sharing. Most readers won’t care about this. Most system builders will.
As things stand right now server blades are often more cost-efficient, smaller and consume less power than traditional servers. But it is likely there is a limit to their scalability. The guidance when looking at blade servers for storage would then seem to be that, at least as the technology stands right now, they are likely a cost-effective alternative for the midrange, but have no way of scaling to the point where they can be expected to challenge high-end arrays.
