Analytical Model for Service Response Time in Distributed Systems

Peer-to-peer (P2P) systems have emerged as a cheap, scalable and self-repairing file sharing, backup/storage and video on-demand (VoD) solutions. Such distributed systems rely on data fragmentation and distributed storage of data over multiple peers. Files are usually partitioned into fixed-size blocks that are themselves partitioned into fragments. In such systems, some peers (clients) exploit the power of parallelism to download blocks/fragments of data stored in a distributed way over some other peers (servers). This idea is deployed even over a network of servers (mirrors and cache) to reduce download time. Determining response times in parallel downloading with capacity constraints on both the client downloads and server uploads is then an important measure of the quality of the service given by these systems. This download time depends on the throughput of the different connections created to obtain the fragments of the given file. This paper enriches the literature with a simple stochastic model of the download process in such large and complex systems. In particular, it characterizes the response time performance metric and assesses its expectation through Markovian model and validates the study through the packet level Network Simulator NS-2. The relative error between the numerical results of the proposed model and the result of the simulation is reported. The comparison shows that the Model characterizes with high accuracy the response time under realistic assumptions.