In Grid systems, users may require assurance for completing their jobs on shared resources. Such guarantees can only be provided by reserving resources in advance. In this paper, we introduce an elastic reservation model, where users can query about a resource availability on a given time interval. They can also provide a reservation duration time and/or number of compute nodes needed as soft constraints to the query. Next, we provide an adapted version of an on-line strip packing algorithm, that takes into a consideration of resource utilization when processing reservation requests. We evaluate our algorithm with a real workload trace and show that the proposed algorithm manages a higher resource utilization and number of acceptance compared to an ad-hoc rigid approach.