Exhibits, Demos & Posters
Storage Space, Advance Reservation and Scheduling
Authors
- Sree Lakshmi D. Gudreddi, Department of Computer Engineering, Santa Clara University
- Silvia Figueira, Department of Computer Engineering, Santa Clara University
Abstract
With the proliferation of data grids, tremendous amounts of data are being generated and stored across widely-distributed servers in the grid. Often, massive amounts of scientific datasets have to be transferred to data centers, which are closer to science centers, for analysis and manipulation. To ensure availability of storage at the destination data centers, space must be reserved in advance.
We defined an advance reservation model to schedule storage space in data centers. Advance reservation guarantees availability of resources at the required time in the future. However, it introduces storage fragmentation resulting in higher blocking probability. We show that using flexibility in the start time of reservations lowers blocking probability and also increases utilization.
The model was assessed using simulation techniques. A trace-generator program was developed to generate traces with reservation requests. Typically, a request is defined by the arrival time, reservation size, duration, and flexibility window size. The random values for the request parameters were generated using stochastic distributions. The scheduler program uses reservation techniques to schedule the request workload generated by the trace generator. The reservation techniques studied were: first-fit, best-fit, worst-fit, and last-fit. The scheduler either accepts, rejects, or blocks each request, according to the current availability at each moment in the simulation.
Experiments were conducted for several traces obtained by varying the request arrival rate and the flexibility limits. We showed that flexibility improves performance. We also showed that, in the scenarios studied, the worst-fit reservation technique is the most effective in both lowering the blocking probability and in increasing the storage space utilization.
