Grid Job Routing Algorithms - Phosphorus

Grid Job Routing Algorithmssmall networks (up to 300 nodes). For this reason, we proposed two heuristic methods to solve this problem[Stevens07a): CP and SP separated and combined optimization. Contrary to the global optimizationperformed by the exact ILP, both heuristics decouple the proxy placement problem from the traffic engineeringbetween the proxies, which results in a two-step optimization plan:(1) Find suitable CP and SP locations and determine which target sites to use;(2) Optimize the flow between CPs and SPs.In fact, step (2) does not contribute to the proxy placement and dimensioning optimization, but allows us toexamine the efficiency of the proxy locations determined in step (1).Average number of proxies2015105Exact solutionCombined heuristicSeparated heuristic00 50 100 150 200Proxy fixed chargeFigure 6.21: Dimensioning of proxy-based anycast architecture: number of proxiesFigure 6.21 and Figure 6.22 summarize the main results of our dimensioning and planning algorithms.Obviously, an increasing fixed charge for installing a proxy (either a CP or SP) leads to less proxies beinginstalled and a growing path stretch. Additionally, the following conclusions can be drawn:1) Both heuristics follow the same trend as the exact optimization, and both provide near-optimal results.2) Separated optimization generally yields results with a smaller path stretch, at the expense of installing moreproxies.Combined optimization suggests a smaller number of proxies and a larger path stretch. On the contrary, thecombined heuristic initially overestimates the infrastructure costs by coupling CP and SP functionality.Afterwards, the infrastructure costs can often be reduced when excess (unused) functionality is removed.Project:PHOSPHORUSDeliverable Number: D.5.3Date of Issue: 31/06/07EC Contract No.: 034115Document Code: Phosphorus-WP5-D5.381