Utility computing SLA management based upon business objectives.

It has become increasingly desirable for companies worldwide to outsource their computing resources, e-business applications, and business processes, to focus on the growth of their core competency and to competitively improve their productivity by exploiting leading-edge computing technologies. Aiming at capitalizing on this information technology (IT) outsourcing trend, leading IT providers are exploring cost-effective means of maximizing the utilization of shareable computing and human resources under the utility computing model. (1) From the customer's viewpoint, the utility computing model promises on demand delivery of IT capabilities and cost-effective usage-based pricing schemes. Service-quality management objectives are assured by the provider in accordance with the established service level agreement (SLA) contract. The customer need not know the implementation details of the provider's service level management (SLM) processes. (2)

A utility computing SLA is an IT service contract that specifies the minimum expectations and obligations that exist between the provider and the customer of a utility computing service. (3,4) It includes one or more service level components, each of which specifies the measurement, evaluation, and reporting criteria for an agreed service-quality standard (5) such as:

* How raw quality measures (e.g., service availability or performance) for an agreed service component (e.g., on demand storage provisioning) in the contract will be gathered

* How raw quality measures will be adjudicated to become qualified quality measures (so that, for example, "service outages caused by the customer or associated with contract maintenance provisions do not contribute to the total service downtime calculations" (6))

* How qualified quality measures will be used to evaluate the achieved service levels (e.g., computing monthly Lotus Notes * availability as "the monthly average availability of the Lotus Notes application running 017 the e-mail servers, weighted by the number of Lotus Notes IDs on each server")

* How service level evaluation results will be reported (e.g., "monthly network latency statistics can be viewed at the following URL [uniform resource locator]")

* How unexpected disputes on service level evaluation results will be resolved

Based upon the agreed set of quality standards (or service level targets), ramifications of not meeting of exceeding the standards can be explicitly included in the SLA contract. If a service level target (or a service level objective (2)) is linked with a penalty clause for a service level violation, it is considered to be a service level guarantee (SLG); otherwise, it is a service level intent. The clarity, attainability, and manageability of a service level guarantee are usually better than those of a service level intent in a commercial SLA contract.

A service level target in an SLA contract can be stated based upon objective quantitative measurement of computing system availability or performance (e.g., "monthly availability of Individual Web Server will be no less than 99. percent") or business process efficiency or effectiveness (e.g., "no less than 97 percent of on demand storage provisioning requests are fulfilled within two business days"). The refund policies for missing service level targets can be specified relative to the service cost (e.g., "credit customer one day of the service cost if the outsourced e-business infrastructure is unavailable more than 15 minutes a day") or in absolute terms (e.g., "credit customer two thousand dollars if a monthly average network latency across the provider ISP [Internet Service Provider] access links to the ISP's backbone is higher than 95 millisecond"). A sample (abridged) Web hosting SLA contract is provided in the Appendix.

>From the viewpoint of a utility computing provider, offering a few customer-neutral service functions atop a common service delivery infrastructure exploits economy of scale better than pursuing a high degree of customization of its service functions for every potential customer. This customer-neutral approach to establishing SLA contracts is adopted by most network and server collocation service providers. (7) However, a competitive IT outsourcing contract normally requires nontrivial customer-specific customization or extension of the provider's "standard" service offerings to accommodate the customer's unique IT outsourcing needs. When the number of such customer-specific SLA contracts grows, the complexity increases in the provider's service delivery environment and SLM processes. (8) A credible study on a leading IT service provider's SLA reporting cost, for example, has shown that several millions of dollars could be saved annually by reducing the cost of generating the monthly reports for 100 high-valued customer-specific SLA contracts by no more than 20 percent. Thus, it is important for a successful utility computing service provider to be able to satisfy its customers' demand for customer-oriented IT outsourcing functions with high-quality services and to fulfill all of its SLA commitments based upon business objectives (e.g., cost-effectively minimizing the exposed business impact of missing SLA commitments).

Figure 1 highlights the business logic for service level reporting (or SLA compliance reporting). It shows that the gathered raw quality measures must be adjudicated first before they can be used as qualified quality measurement. The service level evaluation step can be triggered to generate the quality attainment reporting data for a past (completed) service level evaluation period of for the current evaluation period. After making changes to the input of implementation of any one of the steps, that step as well as the following steps must be re-executed to update the affected service level reports. Activation of such a report update process is necessary when, for example, the qualification status of a quality measure needs to be changed after a dispute about the quality measure is resolved between the customer and the provider.

[FIGURE 1 OMITTED]

In order to make timely adequate service management decisions based upon the provider's SLA commitments, the provider's SLA management system must be capable of performing on demand intermediate service level evaluations with support for adjudication processes for contractual quality measures. The intermediate service level evaluation results must be as accurate and current as possible so that appropriate SLM processes can be executed in a timely manner.

A leading utility computing provider must proactively: (1) maximize customer satisfaction with competitive service level reports (with regard to both coverage and attained quality), (2) minimize the exposed business impact of service level violations, and (3) lower the cost-to-quality ratio of executing SLM processes. However, these goals cannot be sufficiently and effectively supported by the existing service quality management products and common service management practices (9) for the following reasons:

* Existing service quality management products do not support SLA compliance evaluations well because of their limited support of the adjudication processes for quality measures. (10)

* Contractual and internal quality measures on computing system health or performance are usually sent directly to service personnel (who usually manage systems by experience) or system management agents (which usually manage systems by infrequently changed thresholds or condition-action rules). Most service personnel and system management agents know little about the established SLA contracts; moreover, most of them incorrectly equate contractual service level targets to raw quality-monitoring thresholds.

* Service levels on efficiency or effectiveness of business processes (e.g., resource provisioning processes and problem resolution processes) are usually managed by a simple and static task prioritization scheme, such as those based upon severity levels.

* When computing-resource or human-resource contention situations, or both, are caused by unexpected system management alerts, ad hoc SLM processes are usually used to determine which management actions should be carried out first by the available service personnel or system management agents. Resolution-time-based business impact assessments of the alerts are not clearly linked with the provider's SLA commitments and the intermediate service level evaluation results for the affected SLA contracts.

Existing service-quality management technologies and methodologies, therefore, need to be improved to enable unified, business-oriented approaches to fulfilling SLA commitments. (11,12) This paper presents the design rationale of the utility computing SLA management system called SAM (SLA Action Manager) and our implementation experiences with it. The SAM project aims to develop a generic SLA management framework and an integrated set of advanced service level management technologies that among other benefits do the following:

1. Enable the provider to deploy an effective means of capturing and managing SLM-related contractual data as well as the provider's internal management data.

2. Enable the provider and the customer to review and analyze intermediate service level attainment reports on demand.

3. Assist service personnel and service management agents in ordering quality management alerts based upon the exposed business impact over time.

4. Automate the prioritization and execution management of SLM processes, including the assignment of SLM tasks to service personnel using continual optimization technologies.