Distributed computing environments require strategies by which components can detect and recover from failures in remote, collaborating components. Many protocols for distributed systems employ a strategy based on leases, which grant a leaseholder with access to data or services for a limited time (the lease period). If the leaseholder does not renew a lease before expiration of the lease period, the lease grantor assumes the leaseholder has failed and terminates the lease (withdrawing the previously granted access). Choosing an appropriate lease period requires consideration of tradeoffs among resource utilization, responsiveness, and the number of leaseholders. We investigate these issues in the context of Jini Network Technology, a service-discovery protocol created by Sun Microsystems. First, we establish quantitative tradeoffs among lease period, bandwidth utilization, responsiveness, and system size. Then, we consider two self-adaptive algorithms that enable a Jini system, given a fixed allocation of resources, to vary lease periods to achieve the best responsiveness as system size varies. We compare the performance of these self-adaptive algorithms against each other, and against fixed lease periods chosen to accommodate a specific system size. We find that one of the self-adaptive algorithms, based on a simple restriction to the Jini specification, proves easy to implement and performs reasonably well. We anticipate that similar procedures could add self-adaptive capability to other distributed systems that rely on leases.
2003 IEEE International Conference on Pervasive Computing and Communications (PerCom)
, Mills, K.
and Rose, S.
Self-Adaptive Leasing for Jini, 2003 IEEE International Conference on Pervasive Computing and Communications (PerCom), Undefined, [online], https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=50712
(Accessed March 5, 2024)