Supporting SLA provisioning in grids by risk management processes

Grid technologies have reached a high level of development, however core shortcomings have been identified relating to security, trust, and dependability of the Grid which reduce its appeal to potential commercial adopters. Users require a job execution with a desired priority and quality. In order to stipulate such requirements, Service Level Agreements (SLA) can be negotiated. These are a powerful instrument enabling the specification of the business relationships between service providers and service users in detail. However, providers are aware of various threats for SLA violations and are reluctant to adopt a mechanism which requires them to meet strict requirements and to guarantee associated quality constraints. If strict guarantees cannot be agreed by contract, many users prefer to operate their own resources instead of using the Grid. This is more expensive but they control their applications, which removes the issues of trust and ensures dependability concerning its successful completion. To establish a commercial Grid environment, it is essential that Grid providers are prepared to accept an approach involving SLAs with associated guarantees. In order to enable providers to accept such SLAs, they need estimates of the likelihood that they are unable to fulfill an SLA, i. e. Risk Assessment. Furthermore the resource management should take into account such estimations when allocating resources or initiating fault-tolerance mechanisms, i. e. Risk Management. This work integrates risk awareness in the provider’s processes which are involved in SLA provisioning: During SLA negotiation they evaluate which resources can be used for service provisioning and estimate the Probability of Failure (PoF) of resources and of fulfilling the SLA. If the estimated PoF is too high, then, by applying risk reduction mechanisms, the provider may be able to reduce it sufficiently to accept the SLA. The estimated PoF will also be considered by the service provider and service consumer when determining the revenue and the contractual penalty. Compared to a service request requiring a relatively low quality of service, providing a more reliable service requires to receive a higher price since more guarantees have to be ensured. If a more reliable service is provided, the consumer might also define a higher contractual penalty. Thus, the PoF is an additional decision making element in the SLA negotiation since it enables end-users to compare different SLA offers by an objective measurement. When providers have accepted an SLA, they have to be able to compensate for resource failures in order to prevent SLA violations. The usage of fault-tolerance mechanisms combined with risk awareness support Grid providers in this task. The Risk Management processes are interlaced with the resource management and thereby transparent for Grid service consumers. An important aspect of the Risk Management developed for the Grid are self-organising mechanisms, which initiate a fault-tolerance action or a chain of them, in order to manage resource instabilities or resource outages. Decisions are made on the basis of financial considerations, such as the profit margin, the cost for performing fault-tolerance, and the expected profit when executing a job. Taking into account such financial factors is of high importance for commercial Grid providers. In conclusion, the integration of Risk Management in the processes of Grid providers is the initial step towards a risk aware Grid. It will increase transparency, reliability, and trust and provides an objective basis for decision processes in the resource management. Risk Management is integrated to address the SLA negotiation as well as the post-negotiation phase and thereby improves the SLA provisioning process in general.