· by Aldrin Montana
The authors are trying to address throughput and latency performance targets for an entire data center, or at least for entire distributed storage systems within data centers. Part of the problem that the authors see is that data centers, or services such as load balancers, treat storage systems as black boxes and so underutilize the performance capacity of these storage systems.
What other approaches or solutions existed at the time that this work was done? What was wrong with the other approaches or solutions?
What is the authors’ approach or solution? Why is it better than the other approaches or solutions? How does it perform?
The authors propose a multi-layered approach which tags I/O requests with information (a deadline), which disk schedulers use to prioritize I/O requests (deadline queue). Separately, a re-ordering set is proposed to make I/O requests that have been dispatched from the deadline queue efficiently serviced.
Why is this work important?