Grid-Federation Sample Clauses

Grid-Federation a new and emerging grid system which we call as Grid-Federation [▇▇▇▇▇▇ et al. ]. General details about this system can be found in Section 2. Our SLA model considers a collection of computational cluster resources as a contract net [▇▇▇▇▇ 1988]. As jobs arrive, the grid superschedulers undertake one- to-one contract negotiation with the LRMSes managing the concerned resource. The SLA contract negotiation message includes: (i) whether a job can be com- pleted within the specified deadline; and (ii) SLA bid expiration time (maximum amount of time a superscheduler is willing to wait before finalizing the SLA). The SLA bid expiration time methodology we apply here is different from that adopted in the Tycoon system [Lai et al. 2004]. In Tycoon, the SLA bid expiration time at a resource is the same for all the jobs irrespective of their size or deadline. In this case, the total bid-processing delay is directly controlled by the local resource auctioneer. In our model, the superscheduler bids with a SLA bid expiration time proportional to the job’s deadline. The focus is on meeting the job’s SLA requirements, partic- ularly the job’s deadline. The SLA contract negotiation in NASA-Superscheduler and Condor-Flock P2P [Butt et al. 2003] is based on general broadcast and limited broadcast communication mechanism respectively. Hence, these approaches have the following limitations: (i) high network overhead; and (ii) scalability problems. Our time constrained SLA bid-based contract negotiation approach gives LRM- Ses finer control over the resource allocation decision as compared to traditional First-Come-First-Serve (FCFS) approach. Existing superscheduling systems in- cluding NASA-Superscheduler, Condor-Flock P2P, Nimrod-G, Condor-G and Legion- Federation [▇▇▇▇▇▇▇▇ and ▇▇▇▇▇▇▇▇ 1996] assumes every LRMS allocates the re- sources using FCFS scheduling scheme. In this work, we propose a ▇▇▇▇▇▇ backfill- ing LRMS scheduling that focus on maximizing resource owner’s payoff function. In this case, a LRMS maintains a queue of SLA bid requests generated by various superschedulers in the system at a time t. Every SLA bid has an associated expiry time. If the concerned LRMS does not reply within that expiry period, then the SLA request is considered to be expired. Greedy backfilling is based on well known Greedy or Knapsack method [▇▇▇▇▇▇▇ et al. 2001], [▇▇▇▇▇▇ et al. 2001], [▇▇▇▇▇▇ and ▇▇▇▇ 1997]. The LRMSes periodically iterates through the local SLA bids and finalizes th...