280 likes | 418 Views
Topology aggregation and Multi-constraint QoS routing. Presented by Almas Ansari. Flow of the presentation. The Scalability Problem Need for Topology Aggregation Topology Aggregation Schemes Assigning Values to Logical Links Multi-Constrained QoS Routing Conclusions.
E N D
Topology aggregation and Multi-constraint QoS routing Presented by Almas Ansari
Flow of the presentation • The Scalability Problem • Need for Topology Aggregation • Topology Aggregation Schemes • Assigning Values to Logical Links • Multi-Constrained QoS Routing • Conclusions
The Scalability Problem • Routing consists of 2 basic tasks: - collecting network state information - finding a feasible path for a connection based on this information • Topology is usually obtained from a link state protocol like OSPF.
As the network grows larger, it is impossible to broadcast topology to every node because it takes too much space, time and bandwidth. • Ways to deal with this problem: - reducing the no. of topology updates (Goal : deliver as infrequently as possible without affecting routing performance.) - topology aggregation (Goal: reduce the size of the messages without affecting routing) - combining both
Need for Topology Aggregation • TA: very important technique to achieve scalability. • Reduces routing information and thereby routing table sizes by very large magnitude. • Achieved by dividing networks into smaller, manageable routing domains.
Internal details of a domain topology is aggregated before broadcasting. • Inside the domain : complete view • Outside the domain: aggregated view
Aggregated view is used by outside nodes to make routing decisions. • Hence aggregated topologies must be as accurate as possible. • An efficient TA scheme must provide an adequate balance between compaction and accuracy.
Topology Aggregation Schemes • Full Mesh • Single Node • Star • Spanning Tree • ? All schemes suffer from varying degrees of distortion.
Full Mesh • All border nodes connected by logical links. • A logical link has QoS parameters like a physical link. • How to come up with these parameters? • This is still a huge matrix of b(b-1)/2 links. • Does not scale well.
Single Node • One node will represent each routing domain. • It has QoS parameters. • Parameters may be the best, worst or average of all links. • Sometimes the values of the diameter of the graph is used.
Star • Here border nodes are connected via logical links to a virtual nucleus. • Bypasses may be allowed.
Spanning Tree • A spanning tree of all border nodes is created. • To make the representation more accurate, start by including crucial links.
QoS Parameters – How to assign them to logical links? • 2 types of QoS parameters: - link constraint (e.g. bandwidth) - path constraint (e.g. delay) • Additive or Restrictive • Other e.g. : delay jitter, cost etc.
A fundamental step in TA is assigning the QoS parameters to logical link. • Choosing these values correctly is crucial because improper values may lead to rejection of supported calls (under-estimation) or crankback i.e. failure to support an accepted call (over-estimation). • Assigning values to logical links is easier to do when one metric is under consideration • Take best, worst or average values.
It is very difficult to do aggregation with bounded distortion when 2 or more parameters are under consideration. • If a link has 2 parameters b and d, we can find separate optimal paths for each b and d. If we can find a path that maximizes b as well as minimizes d, then a jointly optimal path is found. • A jointly optimal path i.e. that provides better values for all metrics may not exist. • In such cases, other ways to assign values are used.
Conventional approaches to assign values for multi metrics: Single Path Parameters Approach - Decide on the most important parameter - How to decide upon the most important parameter? - Find the best path according to this parameter - Assign values of this path to the logical link
Multiple Path Parameters Best Case Approach - find best path between 2 border for each metric - assign the logical link the best values - aggressive method: since high possibility of crankback Multiple Path Parameters Worst Case Approach - find worst path between 2 border for each metric - assign the logical link the worst values - under estimation method: since high possibility of supported calls not being admitted
QoS Routing • The notion of QoS has been proposed to capture the qualitatively or quantitatively defined performance contract between the service provider and the user application. • QoS routing selects network routes with sufficient resources for the requested QoS parameters. • Goal: satisfying the QoS requirement for every admitted connection.
A QoS routing algorithm may fail to find a feasible path for a connection because: - such a path does not exist - the searching space of a heuristic approach does not cover any existing a feasible path • When this happens the system can either reject the connection or negotiate with the application for a looser QoS constraint.
Multi-Constraint QoS Routing • Multi-Constrained QoS routing deals with finding routes that satisfy multiple independent QoS constraints. • Is NP-Hard • The basic QoS routing problems can be: • Link optimization routing e.g. b optimization routing finding widest route from src to dst.
Link constrained routing e.g. b constrained routing finding a path from src to dst such that b is not less than a certain value on all links. Link optimization problem can be reduced to link constrained problem and then solved by a slightly modified DA or BFA.
Path optimization routing e.g. d optimization routing least delay path • Path constrained routing e.g. d constrained routing d of path below a certain value • These problems can be solved by directly by DA or BFA.
Now consider a link has 2 parameters b and d. • Of this we can have several combinations of routing problems. • E.g. link-constrained path-optimization i.e. To find the least delay path that has a bandwidth constraint • Can be solved by a shortest path algorithm which works on a graph whose links that violate the bandwidth requirement have been pruned.
Other four problem classes are: - link-constrained link-optimization - multi-link constrained - link-constrained path-constrained - path-constrained link-optimization • These are solvable in polynomial time by a modified shortest path algorithm.
Other difficult to solve problem classes are: - path-constrained path-optimization e.g. delay-constrained least-cost routing finding the least cost path with bounded delay. - multi-path constrained e.g. delay-delay jitter constrained finding a path with bounded delay as well as bounded jitter.
If all metrics except one take bounded integer values then the problems are solvable in polynomial time by running the EBFA. • EBFA finds all optimal paths at each node. • Very high space and time complexity.
Limited granularity and Limited path heuristics can be used. • Limited granularity heuristic uses bounded finite range to approximate QoS metrics. Problem can now be solved in polynomial time. • Limited path heuristic limits the no. of optimal paths stored at each node, thereby reducing space complexity.
Conclusions • TA is very important to achieve scalability. • All TA schemes suffer from some distortion. • Multi-Constrained QoS routing is difficult.