Understanding the Behavior of Operator Placement Mechanisms on Large-Scale Networks

Abstract

Operator Placement (OP) mechanism dictates the mapping of compute units, aka. operators on the network infrastructure based on the performance objectives requested by the end-user. The behavior of a placement is strongly influenced by the underlying environment e.g., the workload of queries or performance objectives of a large number of users. Due to this reason, it is not sufficient to build Distributed Complex Event Processing (DCEP) just on a single placement mechanism, but it can benefit from an adaptive use of multiple placement mechanisms. Tcep provides this by (i) supporting key interfaces to develop OP mechanisms, (ii) allowing a dynamic exchange of OP mechanisms based on the user requirements and (iii) facilitating deployment of the operators on a wide range of network infrastructure. In this demonstration1, we (i) provide an understanding on the behavior of OP mechanisms in Tcep on large-scale networks using GENI testbed2 and (ii) show using the transition methodology in Tcep, that different OP mechanisms can be utilized to fulfill the performance requirements of a large number of users.