Opinion

posted on 19 May 2008 06:13

IBM has announced its iDataPlex sideways rack for Web 2.0 and cloud computing with the assoiated extreme compute density requirement. It turns the traditional rack sideways so that it is wider and much the less deep than trad' racks. This gives it much better cooling characteristics and you can fit more of these racks into data centres with far less distance between hot and cold aisles.

The rear wall of such racks no longer needs to be a door and can be a water-cooled heat exchanger. With less depth in the cabinets smaller-diameter cooling fans can be used and these are more efficient and use less power. IBM builds the racks and the inserted Flexnodes (2 or 3U shelves) itself. The servers are dual and quadcore x86 processors running RedHat or Novell SuSE Linux.

IBM builds the things out of industry-standard components and ships the turnkey systems to customer sites. A 2U half width server node can have one x86 server, one PCIe slot, and one 3.5-inch drive or two 2.5-inch drives, SAS or SATA. A full width server node can have a server, two PCIe slots, and two 3.5-inch drives or four 2.5-inch drives.

Customers can have a 3U storage node with twelve 3.5-inch SAS or SATA hot-swap drives.

IBM is pitching the iDataPlex at high-performance computing (HPC) environments now and punting it strongly as a Web 2.0/cloud computing business system as that market develops.

Something is missing.

IBM talks of a partner ecosystem to provide additional products or components for the iDataPlex. One is QLogic. Okay, that's obvious, its Fibre Channel HBAs, but to what? Raw Fibre Channel drives?

There is an Ethernet switch vendor in there as well. Okay, that can provide iSCSI access, but to what? IBM isn't mentioning the SAN word.

Consider; this isn't called the iServerPlex, it's the iDataPlex. Where is the data held? Is it in raw disk drives that the servers treat as their own direct-attached storage? No; that idea is not even that tenable in HPC.

The missing ingredient is a storage array. It's not enough to have a storage node. The system needs a networked storage resource logically separate from the servers but networked to them. (Think about virtualised servers and their need for external networked storage.) It needs a storage array with extreme storage density to complement the server nodes and their extreme compute density.

IBM factory-builds the iDataPlex. There's no going and buying an EMC or NetApp array to fit in the same form factor. It makes a nonsense of IBM's efficient Blue Cloud data centre ideas if compute racks are in iDataPlex form and storage arrays are in trad 19-inch rack form.

It seems highly likely that IBM will deliver its own storage array design for iDataPlex and XIV looks an obvious candidate. Perhaps the Scale-Out File System (SOFS)  has a role to play here.

It's possible, indeed likely, that a stirage announcement is ciming to fill in the missing storage link in the iDataPlex story.

 ^{[Chris Mellor.]}