Density is the name of the game in data centers, letting IT administrators put more of everything in smaller, more efficient spaces. The ultimate in server density is blade servers utilizing SAN storage, which lets IT grow servers and storage without requiring a major increase in physical space. Rack-mounted blade server chassis can provide many powerful servers in a small physical space, to simplify physical management and to economize on power and resources.
In addition, server blades themselves have become progressively more powerful, with a state-of-the-art data center able to provide multiple virtualized servers on each server blade or dedicate blades to serving particularly processor-intensive tasks. But as the utility of blade servers has grown, so have the problems associated with managing them. One of the biggest issues is assigning I/O ports to each blade server for networking and SAN connections.
The problem isn’t so much getting the I/O addresses assigned, which can be done as each blade is brought online, but in making sure that the capabilities, connections, and storage assigned to a blade remain available in case of hardware failure or the simple need to take a server down for some reason. Keeping track of the I/O assignments for a small number of blades isn’t difficult, but when a data center contains dozens or hundreds of blade servers, the I/O management task grows exponentially.
About a year ago, one of the biggest players in the blade server space, HP, introduced the HP Virtual Connect architecture for its c-Class blade server devices (http://h20219.www2.hp.com/services/library/GetPage.aspx?pageid=461242&statusid=0&audienceid=0&ccid=225&langid=121). This architecture is designed to allow virtualization of the LAN and SAN connections assigned to Ethernet and Fibre Channel connections, making them significantly more manageable regardless of the number of physical blade servers or I/O connections and allowing centralized management and control of those connections. By creating a virtualized pool of I/O connections available to the blade servers, HP is able to build a virtualized environment that includes not just server virtualization but also I/O virtualization to networking and storage. Removing the need to physically manage the I/O for each server and virtualizing the I/O connectivity lets IT deliver new services and maintain existing network services with little or no interruption to existing network deliverables. Traditional server provisioning is a time-consuming and complex task that can seriously drain IT resources when large deployments are implemented. But when virtualized, provisioning is quick and simple.
Last month, the other top-tier player in the blade server market, IBM, introduced its BladeCenter Open Fabric Manager (http://www-03.ibm.com/systems/bladecenter/products/openfabric/openfabricmanager.html), which brings I/O virtualization to the IBM product line. Initial marketing has been targeted directly at HP—with claims that the HP solution is product limited and supports only a small selection of the HP products on the market, whereas the IBM solution is fully open and uses standards already supported by I/O vendors in the marketplace. HP has fired back with claims that its architecture is just as open and well supported as the IBM offering. Analysts’ comments on the issue don’t shed much light—they tend to focus more on the “it’s about time” aspect of the IBM technology release.
If you’re a blade server user, or you’re considering introducing blade servers into your data center, I/O virtualization is likely to become a big part of your plans. Take a look at both technologies and see how they each fit into your environment.