Handling HTTP flows over a DiffServ framework

  • Authors:
  • Salvador Alcaraz;Katja Gilly;Carlos Juiz;Ramon Puigjaner

  • Affiliations:
  • Miguel Hernández University, Elche, Spain;Miguel Hernández University, Elche, Spain;University of Balearic Islands, Palma, Spain;University of Balearic Islands, Palma, Spain

  • Venue:
  • Proceedings of the 4th international IFIP/ACM Latin American conference on Networking
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

DiffServ architecture has been widely used to achieve QoS over the Internet. Taking into account that web traffic is the most extended protocol over the Internet community, many solutions have been proposed to supply certain QoS to this type of service. Traditionally, DiffServ architectures have considered two-color markings in order to distinguish between high and low priorities. We study the special treatment for web flows. The web traffic pattern is very close to mouse and elephant distribution flows in Internet. We differentiate flows into short and long classes in order to ensure QoS for short flows, but we try to achieve certain QoS for some long flows. We classify the incoming flows into mice, hybrids and elephants, using three-color markings for each kind of flow, and a three-queue system at the DiffServ system. First of all, we need to detect the gaps in the bandwidth. With these gaps we look for candidate flows to be promoted (hybrids). These hybrid flows will be sent to high priority queue. Next we look for extremely long flows (elephants) which reduce drastically the web traffic performance. Elephant flows will be sent over the low priority queue in order to avoid the promotion of these flows and permit other aspirants to be promoted. Some stochastic functions are computed for the classification at the DiffServ system. Finally, the PLF algorithm is proposed which improves the global performance of the web traffic related with mean of transmission latency and packet loss. We have used ns2 network simulator tool for the simulation with the PackMime-HTTP object for the realistic synthetic web traffic generation.