Replication is known to offer high availability in the presence of failures. This paper considers the case of a client making invocations on a group of replicated servers. It identifies attributes that typically characterise group invocation and replica management, and the options generally available for each attribute. A combination of options on these attributes constitutes a policy. The paper proposes an implementation framework which, by its group-oriented nature, simplifies the task of supporting these policies. It then considers a client (in UCL, London) making invocations on a replica group (in Newcastle, UK) over the Internet. It evaluates the response latencies for four policies that seem appropriate for this set-up. The evaluation takes into account the timing of server crashes with respect to client invocations; both real and virtual failures are considered, the latter being not uncommon in the Internet environment. The experiments are carried out using a CORBA compliant sy...
G. Morgan, P. D. Ezilchelvan