IBM Virtualizes I/O in BladeCenter Servers
December 3, 2007 Timothy Prickett Morgan
The problem with most physical objects, like the gear in a data center, is that it is not malleable. Over time, successive pieces of this infrastructure has been virtualized to make it easier and faster to reconfigure it for different uses, starting with main memory four decades ago and culminating in the wholesale virtualization of entire server stacks over the last decade using virtual machine hypervisors. The problem is, even in a virtualized server environment, those pesky links to networks and storage in the outside world remain physical.
Which is why Hewlett-Packard announced a virtual switch and storage area networking technology for its c-Class blades in March called Virtual Connect I/O. IBM is trying to do one better with a new feature for its BladeCenter systems with a similar technology called Open Fabric Manager.
In plain English, Open Fabric Manager allows for the hard-coded Media Access Control (MAC) address in an Ethernet port and the hard-coded World Wide Name (WWN) address in a storage area network’s Fibre Channel adapter to be virtualized. Inside the BladeCenter chassis, the blade ports come with preconfigured MAC and WWN addresses at the blade and chassis level. Outside Ethernet or SAN networks see these addresses and can hook into the servers and the chassis just as always. But Open Fabric Manager cuts the link between the MAC and WWN addresses the world outside the BladeCenter chassis sees and the blades and switches inside the chassis use such that if you yank out a blade server and plug in a new one in the same slot, that new blade inherits the connectivity of the old blade. This means network administrators do not have to fuss with settings in the Ethernet network or SAN fabrics to match the “physical” addresses stored in MAC and WWN codes in the new Ethernet or Fibre Channel ports on this new blade. Similarly, on the LAN and SAN side of the Open Fabric Manager, upstream links are virtualized, which means this gear can be tweaked and changed independently of the blade servers. Because this code essentially runs in the chassis, it is independent of the operating system running on the blade servers, whether it is Windows, Linux, AIX, or i5/OS, when the latter becomes available on the new Power6-based JS22 blades early next year.
Conceptually, Open Fabric Manager does more or less what HP’s Virtual I/O Connect does, but HP’s approach requires special electronics built into HP’s own switches, while IBM’s approach is done in the blade chassis itself and can virtualize the addresses of any LAN or SAN hardware that is compatible with the BladeCenter. According to Ishan Seghal, director of BladeCenter marketing at IBM, the reason HP took this approach is that it is easier to develop address virtualization inside the switching devices than doing it independently of the devices. It also has the effect of alienating partners who make switches for your blade servers, too.
The basic functionality of Open Fabric Manager, which does LAN and SAN address virtualization, costs $1,499 per chassis. Open Fabric Manager Advanced Upgrade costs an additional $1,999 per chassis, and this is the feature that allows address settings to be migrated from one blade server to another. The two features will be available starting this week, and IBM is offering discounts on the new features ranging from $500 to $1,000 if you buy before the end of the year.