pClock: An Arrival Curve Based Approach for QoS Scheduling in Shared Storage Systems

Ajay Gulati
Rice University
Houston TX-77005
Arif Mechant
HP Labs
Palo Alto, CA
Peter J Varman
Rice University
Houston TX-77005

To Appear in


the International Conference on Measurement and Modeling of Computer Systems

June 12-16, 2007
San Diego, CA

Abstract:  Storage consolidation is becoming an attractive paradigm for data organization because of the economies of sharing and the  ease of centralized management. However, sharing of resources is viable only if applications can be isolated from each other. This work targets the problem of providing performance guarantees to an application  irrespective of the behavior of other workloads. Application
requirements  are represented in terms of  the average throughput, latency and maximum burst size. Most earlier  schemes only do weighted bandwidth allocation; schemes that provide control of latency  either cannot handle bursts or penalize applications for their own prior behavior, such as using spare capacity.

Our algorithm pClock is based on arrival curves that intuitively capture the bandwidth and burst requirements of applications. We show analytically that an application following its arrival curve never misses its deadline. We have implemented pClock both in DiskSim and as a module in the Linux kernel 2.6. Our evaluation shows three important features of pClock: (1) benefits over existing algorithms; (2) efficient performance isolation and burst handling;  and (3) the ability to allocate spare capacity to either speed up some applications or to a background utility, such as backup. pClock can be efficiently implemented in a system without much overhead.

Full Version is available in pdf, ps.
Slides that I used for Pizza talk presentation at Rice.
Note: these were made in a short duration of time. I will be updating this link with my sigmetrics talk slides in June.