Latency Management in Storage Systems

Storage Latency Estimation Descriptors, or SLEDs, are an API that allow applications to understand and take advantage of the dynamic state of a storage system. By accessing data in the file system cache or high-speed storage first, total I/O workloads can be reduced and performance improved. SLEDs report estimated data latency, allowing users, system utilities, and scripts to make file access decisions based on those retrieval time estimates. SLEDs thus can be used to improve individual application performance, reduce system workloads, and improve the user experience with more predictable behavior.

We have modified the Linux 2.2 kernel to support SLEDs, and several Unix utilities and astronomical applications have been modified to use them. As a result, execution times of the Unix utilities when data file sizes exceed the size of the file system buffer cache have been reduced from 50% up to more than an order of magnitude. The astronomical applications incurred 30-50% fewer page faults and reductions in execution time of 10-35%. Performance of applications which use SLEDs also degrade more gracefully as data file size grows.