SmartDataCenter: Disk IO Throttling for Optimal Performance

June 22, 2011 - by badnima

Disk I/O throttling is one of the latest features of SmartDataCenter and enables cloud operators to selectively cap and throttle Disk I/O operations, thereby improving overall system performance.

Examples of throttling can include:

  • Throttling I/O to locally attached disk to prevent one thread/customer/zone from dominating access to the disk pool
  • Limiting CPU utilization to prevent one tenant from dominating all system resources in a fair-share environment

But how does it work? And how can cloud service providers benefit?

SmartOS, the underlying operating system of SmartDataCenter monitors read and write I/O for each SmartMachine. Based on total system latency and configurable priorities, the operating system can throttle each I/O operation by up to 200 microseconds. These short delays allow other zones to access the disks while operations from a heavy-usage zone are throttled. In effect, the system can slow down heavy traffic to ensure other tenants are not adversely affected. This setup allows an operator to minimize latency across all zones instead of optimizing IOPS in a single zone. In addition, a single zone on a lightly-loaded system will be able to use the full I/O bandwidth of the server.

By controlling access to a limited pool of resources through a combination of timing (when), throughput (how much) and latency (how fast), a cloud provider can manage and even guarantee the quality and level of service to its customers. This in turn enables service providers to have smoother and more predictable I/O latency.

To learn more, take a moment to watch a video on Disk I/O throttling featuring Joyent Software Engineer Bill Pijewski discussing the performance and efficiency benefits of the technology.