Open Data Center Alliance Updates "Noisy Neighbor" Controls
The ODCA (Open Data Center Alliance) updated the Input/Output (IO) Controls usage model to revision 1.1 on March 6. As with the five other usage model updates that were announced on the same day, the main change to the usage model is the addition of RFP language to make it easier for IT administrators and cloud service providers to specify how IO is managed in a multi-tenant environment.
You can read my original blog series starting here.
Basically, every IT infrastructure has a bottleneck. Where the bottleneck is located moves depending on a complex interaction of applications, workload, equipment capabilities, and policies that govern how resource contention is handled. The ODCA IO usage model does a decent job of laying out some ground rules for handling contention when the bottleneck rolls around to network and storage IO in a multi-tenant environment.
The ODCA’s RFP language is aggressive, but fairly general. For example, the second RFP clause states, “Solution should provide the ability to control IO for network traffic, fibre channel traffic, and converged fabric traffic. Please state if solution includes all, or partial abilities, and detail any limitations.” The “...detail any limitations” is certainly advantageous for the prospective client and I think this is a good starting point. However, if your organization is really looking at an open data center project, I recommend that you but much more specific here.
To get specific you need to understand your workloads. If you don’t know what IO levels are currently required to run your workloads within acceptable parameters on your on-premise equipment, you won’t know what to ask for from your cloud data center provider. Or, you won’t be able to understand the possible advantages that will be available in a cloud environment that may make deviations from your current normal operations okay.
The ODCA IO Controls usage model rightly seeks to expose controls and service level choices to cloud data center users so that the bad effects of a resource hogging “noisy neighbor” can be seen and dealt with expeditiously.