Virtualization and clustering can be two faces of the same coin.
Computing virtualization is a very hot topic for data center managers. Whether the motivation is higher utilization, reduced management, or business agility, computing virtualization offers compelling possibilities.
Clustering, alternatively, is a strategy for high-performance computing and load balancing. Clustering lets you aggregate resources, delivering virtual “big-iron” performance.
While clustering and virtualization have distinct goals, many data center managers will attempt to pursue both in different parts of their data centers. Fortunately, emerging technologies can combine clustering and virtualization in the same infrastructure.
Data center managers use technologies such as VMware, Xen or user-mode Linux to make one server appear as many distinct “partitions,” each of which is independent and gets a fair “slice” of server resources. This allows them to push server utilization higher than the average of 15% we find in single-purpose servers. For data centers running hundreds of smaller applications, each of which requires a fraction of a server, this avoids the inherent waste and high operational costs of deploying a rack-full of under-utilized servers.
Clustering presents a more proprietary picture - in most cases clustering is done at the application level and is specific to one application - as seen with Oracle’s RAC or Microsoft’s clustering technology for Exchange and SQL Server. Data center managers must dedicate servers to such a cluster, and cannot mix and match the virtualization and clustering technologies.
They can either get fractional servers or aggregates of servers. This means two distinct infrastructures to support these strategies.
A novel approach by start-up company Virtual Iron bridges these two strategies and indicates the playing field for future virtualization and clustering technology competition. Virtual Iron’s product VFe offers a computing virtualization platform for Linux that can span from a fraction of a server to a cluster of up to 16 processors (on one or more physical servers). This approach allows data center managers to use the same orchestration and management software to deploy both the low-demand fractional servers and the high-demand multi-system clusters. More importantly, the VFe software allows data center managers to scale from one scenario to the other.
Of course, there are still many challenges remaining in efficient use of resources through virtualization and clustering. Each of the available technologies will add some overhead to the use of resources. But the bigger picture is that these overheads are trivial compared with opportunities for operational efficiency and business flexibility.