Performance Optimization

VMs Need Capacity to Perform

A VM lacking access to sufficient capacity will quickly develop performance problems. The result is an application slow down, unhappy users and a system adminstrator scrambling to solve the problem.

Without proper capacity planning, performance bottleneck issues are many times solved by throwing hardware resources at the problem and hoping the change solves the problem. But this is not a winning strategy nor one which can solve the problem. Providing sufficient CPU, memory, storage and storage I/O capacity to a VM is more complex than simply throwing resources at the problem since proper capacity planning must occur despite:

• Changes in the VMs performance profile that require more performance to meet transient needs
• Adjacent VMs that share the same resources needing more capacity at unpredictable times
• The movement of VMs between hosts or datastores
• Fluctuations in storage latency and throughput due to network and storage array load changes

Simply throwing hardware at the problem may not be enough. Proper capacity planning is required.

Unfortunately, detecting current capacity bottlenecks and predicting future constraints at the VM, resource pool, cluster or host level is not a trivial task. Monitoring tools only provide immediate feedback of a VMs capacity availability for just a few performance metrics. Worse, these tools tend to provide real time data which is fine for some management tasks, but for capacity management, it is the trending of capacity usage over time which is critical to solving the problem, not the immediate resource usage level.

Real time monitoring tools is not capacity planning and will not easily uncover performance bottlenecks nor predict future issues.

Solving for Performance Bottlenecks

To correctly uncover performance bottlenecks, VKernel's products collect over 20 metrics sampled at a 30 minute interval for all VMs, resource pools, hosts and clusters. These metrics are then fed into a datawarehouse where complex multivariate analysis is performed on this raw information turning it into actionable information for the system administrator. The result is a solution that:

• Finds current and future capacity based performance problems
• Gives actionable recommendations to resolve those problems
• Provides information on storage related bottlenecks [link to storage solution]
• Tells you where to deploy new VMs without causing capacity bottlenecks that will affect performance