%0 Conference Proceedings
%T Parametric-Decomposition Based Request Routing in Content Delivery Networks
%+ Department of Computer Engineering [Istanbul]
%+ MEDIANOVA
%A Bilen, Tuğçe
%A Kurnaz, Dinçer, Salih
%A Sevim, Serkan
%A Canberk, Berk
%Z Part 6: Vehicular and Content Delivery Networks
%< avec comité de lecture
%( Lecture Notes in Computer Science
%B International Conference on Wired/Wireless Internet Communication (WWIC)
%C Boston, MA, United States
%Y Kaushik Roy Chowdhury
%Y Marco Di Felice
%Y Ibrahim Matta
%Y Bo Sheng
%I Springer International Publishing
%3 Wired/Wireless Internet Communications
%V LNCS-10866
%P 323-335
%8 2018-06-18
%D 2018
%R 10.1007/978-3-030-02931-9_26
%K Content Delivery Networks
%K Load balancing
%K Request routing
%K Queuing theory
%Z Computer Science [cs]
%Z Computer Science [cs]/Networking and Internet Architecture [cs.NI]Conference papers
%X Content Delivery Networks (CDNs) enable the rapid web service access by meeting the client requests using the optimal surrogate server located at their nearby. However, the optimal surrogate server can suddenly be overloaded by the spiky characteristics of the high-bandwidth client requests. This accumulates both the drop rates and response times of the client requests. To solve these problems and balance the load on surrogate servers, we propose a Parametric-Decomposition based request routing at the surrogate servers in CDNs. With the Parametric Decomposition method, we combine the high-bandwidth client requests on origin server with our proposed Superposition and Queuing procedures. Then, we split these requests into more than one surrogate server through proposed Splitting and Adjustment procedures. We model the origin and surrogate servers based on G/G/1 queuing system to determine the load status. In case of high congestion on the origin server, we split client requests to the different surrogate servers instead of selecting one. The split sizes of whole content are adjusted by defining a novel splitter index parameter based on the queuing load and waiting time of surrogate servers. The results reveal that the proposed strategy reduces the load on surrogate servers by 42% compared to the conventional approaches. Moreover, the latency and request drops are decreased by 44% and 57% compared to the conventional approaches, respectively.
%G English
%Z TC 6
%Z WG 6.2
%2 https://inria.hal.science/hal-02269731/document
%2 https://inria.hal.science/hal-02269731/file/470666_1_En_26_Chapter.pdf
%L hal-02269731
%U https://inria.hal.science/hal-02269731
%~ IFIP-LNCS
%~ IFIP
%~ IFIP-TC
%~ IFIP-TC6
%~ IFIP-WG6-2
%~ IFIP-WWIC
%~ IFIP-LNCS-10866