Available IO Capacity-Based Workload Deployment in OpenStack Environments

When it comes to deploying their critical workloads on OpenStack environments a key aspect that enterprises must address is ensuring that the application has adequate resources available to execute well and meet its performance SLAs. 

Of course, Nova scheduling and placement already helps address this by taking required and available memory, storage and CPU cores into account. But while it also takes account of available disk space and concurrent IOPs it does not consider the underlying storage hardware that provides the IO capacity or the burstiness that certain workloads may present. 

The risk with not taking account of the available, or residual, IOPs capacity is that the deployment decisions can be less than ideal potentially resulting in workloads being placed on nodes unable to meet performance SLA requirements. 

Veritas is adopting an approach augments Nova’s filters and takes account of available IOPs across the compute plane in deciding about workload placement. This has important benefits:

  • Improved VM packing density and corresponding storage utilization
  • Improved ability to meet application SLAs through mitigating noisy neighbor interference
  • Ability to dynamically load balance taking account of IO load

You can watch my engineering colleagues discussing this approach during the vBrownBag TechTalks at OpenStack Summit in Barcelona.