2010 Prediction: Varun Nagaraj, Aprius

By Varun Nagaraj
published: Tuesday, December 08 2009

Writing a piece on how the future will unfold is a risky proposition.  If it stretches too far, the response could be “Can you pass on that good stuff?”  If you play it too safe, the reaction could be a biting and obvious, “You think?”  With that in mind, let’s take a look at some incremental and disruptive predictions.

The incremental predictions: Having tasted success with server virtualization, data center managers will look to double or triple the average number of virtual machines (VMs) per host from less than 10 today to nearly 20 by the end of 2010.  So far, data center managers have left many of the heavier database or messaging applications untouched and not incorporated them into their virtualized server farms due to concerns regarding performance and I/O bottlenecks.  This will change in 2010.  Data center managers will begin to incorporate technologies to support greater VM density and heavier workloads.  On the processor side, the new generation of multi-core machines will step up and deliver.  On the I/O side, the data center manager will have a range of options to provide the required connectivity – from over-provisioning many and large physical network connections for each server to implementing emerging technologies such as I/O Virtualization that can provide consolidated, cost-efficient and flexible I/O to the hosts.

The disruptive predictions:  In spite of the use of the term “server virtualization,” it is the application that has been the primary focus of virtualization so far.  Turning applications into VMs creates portable applications that can be deployed and moved around on a physical infrastructure.  A VM can describe the physical environment it needs – how much CPU, memory, disk, network connectivity and so on.  The assumption is that when a VM lands on a host, it can get the physical resources it needs.  But when each host is defined as a statically pre-provisioned bundle of hardware attributes, this mapping is far from easy.  In 2010, data center managers will look to provide a flexible infrastructure that can provide a VM with the environment it needs, no matter where the VM lands.  This concept of abstracting the infrastructure can be described as infrastructure virtualization.  The goal of infrastructure virtualization will be to think of the data center as a collection of CPU, memory and I/O pools that can be dynamically put together to support the needs of any VM.  Today’s rack servers and blade servers will provide the CPU and memory building blocks.  The I/O pools and the binding of the I/O to the rack or blade servers will be provided by a new class of product called the “I/O Gateway.”  The unit of infrastructure deployment and management will no longer be the individual rack server or blade server; it will be a “pod” that includes pools of CPU, memory, and I/O, and that provides an environment for easy VM mobility.