In this paper, we present an optimization model for storage service based on quality of service. The quality measure is the sojourn time of an individual I/O request. The service provider gets a premium P for each I/O request served, commits to serve certain I/O request rate lambda, and in return, the service provider guarantees its service quantitatively using a rebate model. Intuitively, the service guarantee is quantified as a rebate to the customer depending on how well the service is rendered. The overall objective for the storage service provider is to find an optimal I/O rate it should commit to serve to maximize its net gain, that is, the difference between the premium received and the rebate back to the customer.
This model has a clear and quantifiable measure of net gain. The provider has every reason not to over-commit its service capacity while the customer has a way to verify the guaranteed quality of service in monetary terms. To gain insight on the structure of the model, we show some analytic results for this optimization problem with a simple M/M/1 FCFS queuing model. Two scenarios are examined. One uses queuing time of the I/O request as the measure of quality of service and the other uses response time.
We have a few suggestions on how to apply this model in the real-life world. To operate at an optimal level, storage service providers have to quantitatively characterize the customer workload and their own storage subsystems. On the operational side, the storage service providers need to continuously
monitor and shape the arrival stream for conformance and deliver the guaranteed level of service. Postmortem analysis of I/O traces and measurements should help in determining and projecting sustainable and profitable I/O rate.
We could anticipate the existence of a third-party measuring and reporting agent which is between customer applications and storage service provider. With this arrangement, full accountability shall prevail, and the proposed model would make even better business sense.
By: T. Paul Lee
Published in: RJ10242 in 2002
LIMITED DISTRIBUTION NOTICE:
This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.
Questions about this service can be mailed to reports@us.ibm.com .