Browse Prior Art Database

Dynamic implementation switching to support service level agreements on demand Disclosure Number: IPCOM000028699D
Original Publication Date: 2004-May-27
Included in the Prior Art Database: 2004-May-27
Document File: 2 page(s) / 65K

Publishing Venue



One of the problems when moving to a utility model of services such as in on demand computing is that the service must publish and support some level of service to its users. This service level agreement (SLA) will contain many aspects of the service being provided but one of these will be performance. For example, what kind of response time can a caller expect from this service. Due to the on demand nature of these kinds of services the calling frequency, kinds and amount of data and the number of concurrent calls is likely not known in advance. This means that handling these variations at runtime will be required to ensure that the SLAs are still met. There are many solutions that attempt to adjust to the dynamic environment of an on demand solution which revolve around the dynamic addition or tuning of resources to support service level agreements. However, the novel approach with this invention is the realization that in some cases an entirely different implementation might be required for supporting the service level agreement depending on many factors. This invention describes a mechanism for the service to evaluate these factors and switch the implementation to the appropriate one to meet the service level agreement. By providing multiple implementations, a better level of service can be guaranteed than if the service was implemented with a single implementation which had to handle all situations.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2


CA820030136 Monica Rooney/Markham/IBM Mike Starkey

Dynamic implementation switching to support service level agreements on demand

This article describes a way of handling the dynamic nature of on demand, while supporting a service level agreement, by providing a mechanism to have multiple implementations of the service that provides the same results. The different implementations have different characteristics and make different tradeoffs - one may use more resources but take less time, another might be slower but require fewer resources. Other solutions to this problem look at tailoring resources and execution characteristics of a single implementation to meet a service level agreement. However, in some cases this will be difficult to do and the algorithm may be suboptimal for certain data or environments regardless of the amount of tuning done. This article describes a mechanism to provide the optimal implementation for the current environment, dynamically and transparent from the caller.

First, there can be multiple implementations that may produce the same solution which have different characteristics and make different tradeoffs. Figure 1 shows the response time of a service implemented two different ways (QS and BS). For certain sets of data, BS is the faster implementation. However, at some point (around 17 input data in this case), QS becomes the faster implementation. With the service dynamically switching its implementation when this boundary is crossed, the fastest response time can be achi...