In any virtualized computing environment there is a very fine balancing act to carry out.
On the one hand you have to make sure there is enough infrastructure capacity to meet the resource demands of application workloads, so performance doesn't take a hit. But on the other hand you want to avoid overprovisioning, because that's an inefficient use of resources and therefore a way for money to disappear down the drain.
Getting that balance right is what VMTurbo calls the "Desired State." VMTurbo is a company that we've looked at several times, most recently when VMTurbo released its Operations Manager 4.0 application. The software uses market economics to allocate VMs to resources — resources in your data center, perhaps in a private cloud, as well as resources available in public clouds like Amazon's or Azure's.
These guys know a thing or two about optimizing application workload performance and infrastructure efficiency, so it's worth taking a look at their current thinking.
There are four key aspects to look at that can help enterprises strike a balance, according to VMTurbo.
Intelligent Workload Placement Decisions Within Clusters
Within a server virtualization cluster, workloads consume resources such as CPU, memory and network I/O, and the amount of resources they consume fluctuates over time. Simple scheduling mechanisms offered by the likes of Microsoft and VMware monitor utilization and shunt the biggest resource hogs over to the least-used hosts in a cluster to equalize things out within the cluster.
But VMTurbo contends that a more holistic approach can be more effective in achieving the Desired State. The company reckons that by examining how applications workloads could be best placed in the environment to preserve the quality of service — preventing workload interference whilst maximizing the efficiency of the infrastructure — it is possible to achieve efficiency gains of 20% - 40% over native hypervisor scheduling.
Intelligent Workload Placement Across Clusters
Software-defined networking (SDN) has made it much easier for companies to move their VMs around, and virtualization technologies like vMotion and Live Migration provide the ability to move working VMs between clusters. But VMTurbo points out that what's generally missing is the capability to analyze cross-cluster workload placement decisions on a continuous basis to move the virtualized server infrastructure towards the Desired State.
"Intelligent Cross Cluster workload placement opens up a whole new opportunity to better exploit islands of compute resource, which may be underutilized," the company says. "Not only does this enable more efficient use of resources by eliminating the need to provision for peaks in cluster utilization, it also provides a safety net to accommodate unplanned peaks in application workload resource utilization," it adds.
VMTurbo contends that many organizations have built up their server virtualization hardware infrastructure around specific projects, and in many organizations this results in smaller clusters of resources that are underutilized. A better solution would be to consolidate these resources into larger clusters.
What's needed, according to the company, is the ability to simulate the potential impact of consolidation where larger pools of resources are built from existing hardware. This can result in significant cost savings by increasing underutilized resources, the company believes.
Recovering Unused Reserved Server Capacity
VMware and Hyper-V provide a way to reserve resources for individual workloads to guarantee that the resources will be available to an application — even when they are not needed — to mitigate performance issues.
VMTurbo points out that this is inefficient because inevitably more capacity is reserved from the underlying infrastructure than is actually needed. Virtual machine rightsizing — the ability to set reservations based on the actual amount of resources that are required by individual workloads — can result in the recovery of significant memory and CPU resources, resulting in increased hardware utilization and cost deferral, the company points out.
So what's the point of all this? Rather than try and carry out these balancing acts — workload placement within and between clusters, resource consolidation and VM rightsizing — by hand or by using the rudimentary tools provided by the hypervisor makers, it may be better to use specialist software tools.
Of course, VMTurbo is going to tout that as the case, because it sells resource allocations software. Still, if software-driven control can increase server virtualization infrastructure efficiency by as much as 40%, as the company claims, then it's got to be worth taking a look at.
Paul Rubens is a technology journalist and contributor to ServerWatch, EnterpriseNetworkingPlanet and EnterpriseMobileToday. He has also covered technology for international newspapers and magazines including The Economist and The Financial Times since 1991.