On synchrony in dynamic distributed systems

Abstract

Abstract Many modern distributed services are deployed in dynamic systems. Cloud services are an example. They are expected to provide service to a potentially huge amount of users and may require a wide geographical deployment in multiple data centres. Their service processes vary in volume in accordance with workload variations, showing an adaptive behaviour in order to minimise economical costs. Dynamic distributed systems may be classifed considering two axes: (a) the number of processes that compose the system, and (b) the diameter of the networking graph that interconnects those processes. Other important features of dynamic systems can be derived from these two characteristics, e.g., their attainable synchrony. We analyse the level of synchrony that may be achieved in each dynamic system class and revise the existing techniques for transforming an initially asynchronous large dynamic system into another one with a higher synchrony level. With this, a larger set of problems may be handled in dynamic distributed systems. This facilitates the implementation and provision of additional services in those systems.