Object replication strategies in content distribution networks

Comparison and analysis of ten static heuristics-based Internet data replication techniques

“…In the context of the data replication problem, the capacity-free (unlimited storage) version has a better worstcase performance than the capacity-constrained version [42], yet it requires a lot of maneuverability in terms of choosing the optimization function [9]. In [30], the authors use the capacityconstrained version of the minimum k-median problem and guarantee a stable performance. However, such results are possible only with highly conservative assumptions (such as, fixed location of the original server, access patterns are to be known beforehand, no network failures, etc.)…”

“…This approach has many advantages [51], such as it saves the server memory capacity, moves only those objects that are actually required to be reallocated, and provides load-balancing. The generalized fine-grained replication is known to be NP-complete not only for general graphs [41], but also for partitioned graphs [30]. The work proposed in [41] analyzed both static (such as, a modified Greedy based approach [44] and an Evolutionary method based on Genetic algorithms [23]) and adaptive (such as, a self-configured Genetic approach) replication techniques.…”

Content Distribution for Publish/Subscribe Services

“…Kangasharju et al [18] propose four heuristics, but the best one needs global knowledge about the network topology, the reference distribution and the content distribution at different times. Qiu et al [25] propose several heuristics to choose M replica sites from N candidates for a given site, assuming a relatively stable reference pattern at the candidate sites.…”

Caching Techniques on CDN Simulated Frameworks

“…The content of the static cache is identified by applying a content replication algorithm. A wide range of content replication algorithms have been proposed in literature [12,19,21,32,37]. Kangasharju et al [12] use four heuristic methods: 1) random, 2) popularity, 3) greedy-single, and finally 4) greedy-global.…”

“…Consequently, the necessity of a simulation environment for performing experiments, still remains. Towards this direction, there have been several implementations of a simulated CDN [4,7,12,34] which fit the individual needs of each research work. Most of them do not take several critical factors into account, such as the bottlenecks that are likely to occur in the network, the number of sessions that each network element can serve (e.g.…”