Everest-OSDI08 - Everest: Scaling down peak loads through...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Everest: Scaling down peak loads through I/O off-loading Dushyanth Narayanan Austin Donnelly Eno Thereska Sameh Elnikety Antony Rowstron Microsoft Research Cambridge, United Kingdom { dnarayan,austind,etheres,samehe,antr } @microsoft.com Abstract Bursts in data center workloads are a real problem for storage subsystems. Data volumes can experience peak I/O request rates that are over an order of magnitude higher than average load. This requires significant over- provisioning, and often still results in significant I/O re- quest latency during peaks. In order to address this problem we propose Everest, which allows data written to an overloaded volume to be temporarily off-loaded into a short-term virtual store. Everest creates the short-term store by opportunistically pooling underutilized storage resources either on a server or across servers within the data center. Writes are tem- porarily off-loaded from overloaded volumes to lightly loaded volumes, thereby reducing the I/O load on the for- mer. Everest is transparent to and usable by unmodified applications, and does not change the persistence or con- sistency of the storage system. We evaluate Everest using traces from a production Exchange mail server as well as other benchmarks: our results show a 1.470 times re- duction in mean response times during peaks. 1 Introduction Many server I/O workloads are bursty, characterized as having peak I/O loads significantly higher than the av- erage load. If the storage subsystem is not provisioned for its peak load, its performance during peaks degrades significantly, resulting in I/O operations having signifi- cant latency. We observe that workloads are usually un- balanced across servers in a data center, and often even across the data volumes associated with a single server. We propose Everest, a system that improves the perfor- mance of overloaded volumes by transparently exploit- ing statistical multiplexing across the storage bandwidth resources in the data center. Everest monitors the performance of a data volume, and if the load on the volume increases beyond a pre- defined threshold, it utilizes spare bandwidth on other storage volumes to absorb writes performed to the over- loaded volume. It does this by maintaining a virtual short-term persistent store , into which data is temporar- ily written, or off-loaded. The store is virtual in the sense that storage resources are not explicitly allocated to it; rather it is created by pooling idle bandwidth and spare capacity on existing data volumes either on a sin- gle server or across a set of servers in the same data cen- ter. In the common case, this can remove the majority of writes from the peak load, allowing the data volume un- der stress to serve mostly reads. When the peak subsides, the off-loaded data is lazily reclaimed back to the original volume, freeing the space in the Everest store. Everest handles short-term peaks and is not designed to handle long-term changes in load: these must be addressed by...
View Full Document

Page1 / 14

Everest-OSDI08 - Everest: Scaling down peak loads through...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online