Cycle-harvesting software on commodity computers is available from a number of companies and a significant part of the Grid computing landscape. However, creating commercial service contracts based on resources made available by cycle-harvesting is a significant challenge for two reasons. Firstly, the characteristics of the harvested resources are inherently stochastic. Secondly, in a commercial environment, purchasers can expect the providers of such contracts to optimize against the quality of service (QoS) definitions provided. These challenges have been successfully met in conventional commodities, e.g. Random Length Lumber, traded on financial exchanges and we draw inspiration from there. The essential point for creating commercially valuable QoS definitions is to guarantee a set of statistical parameters for each and every contract instance. In statistical terms this is the difference between guaranteeing the properties of what is delivered versus the source from which the delivery will be made. In this paper we describe an appropriate QoS definition, Hard Statistical QoS (HSQ), and show how this can be implemented for cycle-harvested resources using a hybrid stochastic-deterministic system. We present an architecture and algorithms to support HSQ contracts. We analyze algorithm behavior analytically using a distribution-free approach versus the expected proportion of deterministic resources required for an HSQ specification. For example, where slot lengths are log-Normally distributed we find that for hard guarantees on 8 quantiles with contract sizes 16 to 1024 slots, from 13% to 1% deterministic resources are required. Permitting oversampling is relatively inefficient leading to up to 61% of the stochastic resources being wasted in a typical case. Including downwards substitution reduces deterministic resource requirements by roughly half. We conclude that commercial service contracts based on cycle-harvested resources are viable both from
a conceptual point of view and quantitatively for contracts of sufficient size.
By: Chris Kenyon and Giorgos Cheliotis
Published in: RZ3461 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 .