Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.
September 30, 2005
3 Min Read
Utility computing holds great appeal. The ability to connect servers, networks, and storage systems; virtualize the bandwidth, computing horsepower, and storage arrays into a single resource pool; and dynamically shift those resources to where they're needed promises to substantially improve utilization rates, reduce the number of systems a business needs, and cut costs.
But there's been a big problem with this emerging approach to data-center architecture: It has been pushed by server vendors and, for the most part, their technology only works with their own equipment. Cisco Systems, the world's largest supplier of networking equipment, is pitching a plan to change that. Last week, it introduced a network-centric utility-computing platform that will let businesses mix and match resources from various hardware vendors, including Cisco's rivals.
"We think it's an important and interesting technology advancement that can further utility computing," says Matthew Wolf, a research scientist at the College of Computing at Georgia Institute of Technology, which is testing the Cisco technology to connect a cluster of about 450 Xeon processor nodes within a Dell server installation. Being able to more easily implement virtualization technologies across the cluster will be a significant enhancement. "The jury is still out, and I don't think it's any kind of panacea."
Cisco's Data Center Networking Architecture portfolio is built around two primary elements: an InfiniBand switch, which is designed to accelerate server speed, and Cisco's VFrame virtualization software, which lets users make systemwide allocations of computing and storage resources to deliver a computing framework where resources can be shifted to the jobs that most need them.
The Server Fabric Switch line uses InfiniBand technology to create a unified fabric for connecting large numbers of servers into grids. The Server Fabric Switch can be coupled with Ethernet and Fibre Channel gateway technology to connect the server grids with shared LAN and storage-area-network resources utilizing Cisco's flagship Catalyst switches or MDS 9000 storage-networking switches.
InfiniBand implementations are most prominently used in server clusters where high bandwidth and low latency are key requirements. The InfiniBand standard has been promoted by a variety of IT vendors for several years, including Agilent, Dell, Hewlett-Packard, IBM, Intel, Sun Microsystems, and Topspin Communications, a switch-fabric specialist acquired in May by Cisco for about $250 million.
Cisco is layering its software on top of the InfiniBand switch to provide virtualization, scheduling, and provisioning services to address multiple resources, including switching, data network, load balancing, and security products. VFrame will offer customers a single interface to provision infrastructure elements, rather than hav- ing to address each type of technology individually.
The data-center technology fits into Cisco CEO John Chambers' vision of a "network of networks" based on virtualization of services and resources and the creation of application-aware platforms. "It has huge implications," Chambers said in a May interview. "It makes it really transparent where the application resides or what combination of devices it runs across."
A big unknown is how well the Cisco framework will work with existing utility computing initiatives from server vendors like Hewlett-Packard, IBM, and Sun. The architecture probably will work well with Sun products since Cisco and Sun are key infrastructure partners in the EDS-led Agility Alliance, a multivendor utility-computing group aimed at competing with IBM and HP. EDS says it will implement the Cisco architecture to deliver its own virtual server services to its customers.
You May Also Like