1 / 24

ISP and Egress Path Selection for Multihomed Networks

This paper discusses the selection of upstream ISPs and the allocation of egress traffic in multihomed networks to achieve cost effectiveness and performance optimization.

cporter
Download Presentation

ISP and Egress Path Selection for Multihomed Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. ISP and Egress Path Selection for Multihomed Networks Amogh Dhamdhere Constantine Dovrolis (amogh,dovrolis)@cc.gatech.edu Networking and Telecommunications Group College of Computing Georgia Tech

  2. Multihoming • Multihoming: Connection of a stub network to multiple ISPs • 70% of stub networks are multihomed • Redundancy • primary/backup relationships • Load Balancing • Distribute outgoing traffic among ISPs • Cost Effectiveness • Lower cost ISP for bulk traffic, higher cost ISP for performance-sensitive traffic • Performance • Intelligent Route Control Amogh Dhamdhere IEEE Infocom 2006

  3. Major Questions • How to select the set of upstream ISPs ? • Low monetary cost • Good performance (low delay, loss rate) • Path diversity to major traffic destinations – improves robustness to network failures • How to allocate egress traffic to the set of selected ISPs ? • Objective: Avoid congestion on the upstream paths • Also maintain low cost Amogh Dhamdhere IEEE Infocom 2006

  4. Related Work • Wang et al. (Infocom 2004) • ISP selection to minimize cost to subscriber • Did not consider performance constraints and path diversity • Applicable only to percentile based charging pricing function • IRC systems (RouteScience, Internap, Radware..) • Path switching for better performance • Work on short timescales – Can lead to oscillations • Goldenberg et al. (Sigcomm 2004) • IRC algorithm for optimizing latency and cost over short timescales • Applicable to the percentile based charging model Amogh Dhamdhere IEEE Infocom 2006

  5. Problem Definition • Two phase problem • Phase I – ISP Selection: • Select K upstream ISPs • K depends on monetary and performance constraints • “Static” operation • Change only when major changes in the traffic destinations or ISP pricing • Phase II – Egress Path Selection • Allocate egress traffic to selected ISPs • Avoid long term congestion and minimize cost • “Semi-static” operation, performed every few hours or days Amogh Dhamdhere IEEE Infocom 2006

  6. Assumptions • We provision only the egress traffic of S • The set of M major destinations is known • Average rates to the major destinations are known • Number of ISPs to choose (K) and the set of possible ISPs (I) is known • An ISP charges based on the volume of traffic routed through it (volume based charging) • Assume increasing and concave pricing functions Amogh Dhamdhere IEEE Infocom 2006

  7. Objectives of ISP selection • ISP selection should consider both monetary cost and performance • Minimum monetary cost • Estimate the cost that “would be” incurred if a set of ISPs was selected • Minimum AS-level path lengths • Longer paths can lead to larger delays and increase vulnerability to inter-domain routing failures • AS-level paths can be measured offline using Looking Glass Servers • Maximum Path diversity • AS-level paths to destinations should be as “different” as possible Amogh Dhamdhere IEEE Infocom 2006

  8. ISP Selection • K ISPs to be selected out of |I| • Associate a cost with each performance metric • Monetary cost, path length cost and path diversitycost • Total cost of a selection of ISPs C: ct(C): = αmcm(C) + αpcp(C) + αdcd(C) • Optimization problem: Find the set C* with the minimum total cost • Brute Force approach is feasible • E.g. For |I|=15 and K=4, there are 1365 combinations • Solution approach: Evaluate the cost of each selection and choose the set with the minimum cost Amogh Dhamdhere IEEE Infocom 2006

  9. Monetary and Path Length Cost • Lower level optimization problem: Given a set of ISPs C, what is the minimum monetary and path length cost of routing egress flows ? • Find the mapping G* of items to bins that minimizes the cost of the assignment (Bin Packing) • Flows = items • ISPs = bins • To find: Least cost assignment of flows to ISPs • NP hard ! • Use First Fit Decreasing (FFD) heuristic • Generated mapping G* very close to optimal • Monetary and path length costs of C are then calculated using the mapping G* Amogh Dhamdhere IEEE Infocom 2006

  10. Path Diversity Cost • A selection C gives K paths to each destination d • K-shared link to d: A link which is shared by all K paths to d • If a K-shared link fails, destination d is unreachable • Minimize the number of K-shared links • Should give best performance for single-link failures • Define metric k(d,C): number of K-shared links to d in selection C • Choose the selection with the minimum k(d,C) averaged over all destinations Amogh Dhamdhere IEEE Infocom 2006

  11. Evaluation - Bin Packing • FFD heuristic used to find the minimum monetary and path length costs • Simulations • Need exhaustive search to identify optimal cost • Restrict network to 3 ISPs and 15 destinations • FFD heuristic finds a solution with high probability, when average load is below 60-70% • In high load conditions, the probability of finding a solution decreases • Cost ratio is close to 1, even at high load conditions • FFD heuristic is close to the optimal in terms of cost Amogh Dhamdhere IEEE Infocom 2006

  12. Evaluation – Path Diversity • AS-level paths and traffic rates are input to simulator • 9 ISPs, 250 destinations • Given K, find the selection C* with the minimum path diversity cost • For each selection C, u(C) = total traffic lost due to the failure of each link in topology • Calculate Δu(C) = u(C) – u(C*) for each selection C Single link failures: C* is the optimal selection 2,3 link failures: C* is close to the optimal selection Amogh Dhamdhere IEEE Infocom 2006

  13. Egress Path Selection • After Phase-I, S has K upstream ISPs • Problem: How to map outgoing traffic to the ISPs • M flows: KM mappings of flows to ISPs • Some mappings may cause congestion to flows ! • Flows can be congested at access links or further upstream • Objective: Find the loss-free mapping with the minimum cost • Challenges: • Upstream topology and capacities are unknown • Cannot know a priori whether a mapping will cause congestion • Iterative routing approaches required Amogh Dhamdhere IEEE Infocom 2006

  14. Egress Path Selection • Step 1: Use the FFD heuristic to map flows to ISPs • Assume initially that the access links are bottlenecks • Access capacities are known • FFD heuristic for bin packing gives a cost close to the best possible cost • Some flows may be congested • Bottlenecks in upstream networks • Use as the starting point for the stochastic search Amogh Dhamdhere IEEE Infocom 2006

  15. Egress Path Selection • Step 2: Use iterative stochastic search to find a loss-free solution • Stochastic search by simulated annealing • Iterative combinatorial optimization algorithm • Route traffic, measure congestion, decide next action • Action: Flows re-routed from one ISP to another • can accept moves which increase congestion • Accepting “bad” moves can help to escape local minima Amogh Dhamdhere IEEE Infocom 2006

  16. Evaluation of Stochastic Search • Stochastic search involves iterative routing • Traffic has to be re-routed • Some traffic may be dropped due to congested links • Evaluation metrics • Probability of finding a solution (high) • Number of iterations to find a solution (low) • Amount of traffic re-routed (low) • Amount of dropped traffic (low) • Compare against other heuristics • Only bin packing (access-link) • Greedy iterative algorithm (greedy-single) • Moves which increase congestion are not accepted • Variants of Simulated Annealing Amogh Dhamdhere IEEE Infocom 2006

  17. Evaluation of Stochastic Search • SA-slow and greedy show similar probability of finding a solution • Other algorithms have a significantly lower probability of finding a solution • On the average, SA-slow needs fewer iterations to find a feasible solution • Accepting “worse” solutions can actually help find a loss-free solution faster • SA drops less traffic on the average than greedy-single • SA re-routes less traffic on the average than greedy-single Amogh Dhamdhere IEEE Infocom 2006

  18. Summary of Contributions • Proposed practical algorithms for ISP selection and egress traffic allocation among selected ISPs • ISP selection algorithm takes into account both monetary and performance constraints • Formulated as a bin-packing problem • Applicable for general pricing functions • Can be extended to incorporate more performance metrics • Egress path selection without knowledge of upstream topology • Proposed simulated annealing for stochastic search • Performs better than other simple iterative algorithms Amogh Dhamdhere IEEE Infocom 2006

  19. Thank You ! Amogh Dhamdhere IEEE Infocom 2006

  20. Evaluation of Bin Packing • Simulations • Need exhaustive search to identify optimal cost • Restrict network to 3 ISPs and 15 destinations • FFD-like heuristic finds a solution with high probability, when average load is below 60-70% • In high load conditions, the probability of finding a solution decreases • Cost ratio is very close to 1, even at high load conditions • Heuristic algorithm is close to the optimal in terms of cost Amogh Dhamdhere IEEE Infocom 2006

  21. Stochastic Search – Probability of finding a solution • bneck_loc=0.5 (bottlenecks in the middle of network) and bneck_shar=0 (no shared bottlenecks) • SA-slow and greedy-single show similar probability of finding a solution • Access-link, greedy-mult and SA-fast show significantly lower probability of finding a solution • Henceforth, compare only greedy-single and SA-slow Amogh Dhamdhere IEEE Infocom 2006

  22. Stochastic Search – Number of Iterations • How many iterations before solution is found ? • Number of iterations required increases with offered load (average flow rate) • SA-slow performs better than greedy-single on the average • Accepting solutions with increasing congestion can actually help find a non-congested solution quicker Amogh Dhamdhere IEEE Infocom 2006

  23. Stochastic Search – Results • What is the total rate of traffic that is re-routed ? • What is the total rate of traffic that is dropped due to congestion ? • SA-slow drops less traffic on the average than greedy-single • SA-slow re-routes less traffic on the average than greedy-single Amogh Dhamdhere IEEE Infocom 2006

  24. Assumptions • Provision the egress traffic of S • The set of M major destinations is known • Average rates to the major destinations are known • Number of ISPs to choose (K) and the set of possible ISPs (I) is known • An ISP charges based on the volume of traffic routed through it (volume based charging) • Assume increasing and concave pricing functions Amogh Dhamdhere IEEE Infocom 2006

More Related