1 / 30

Lecture XVII: Distributed Systems Algorithms Inspired by Biology

Lecture XVII: Distributed Systems Algorithms Inspired by Biology. CMPT 401 Summer 2007 Dr. Alexandra Fedorova. Problem Statement. Load balancing in telecommunication networks Calls originate and end nodes and are destined to end nodes

maylin
Download Presentation

Lecture XVII: Distributed Systems Algorithms Inspired by Biology

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. Lecture XVII: Distributed Systems Algorithms Inspired by Biology CMPT 401 Summer 2007 Dr. Alexandra Fedorova

  2. Problem Statement • Load balancing in telecommunication networks • Calls originate and end nodes and are destined to end nodes • Calls are routed through intermediate switching stations or nodes • Each node has a certain capacity – can support only a limited number of calls routed through it • Many routes for each call • Routing tables determine the route • If the call is routed via a congested node, it must be dropped • Goal: construct routing tables that minimize the number of dropped calls under changing load conditions

  3. Potential Solutions • Central controller: knows about the entire system, updates routing tables at nodes • Nodes must communicate with the controller • The controller is a single point of failure • Use shortest-path routing • Determine the shortest path from each source to each destination • Construct routing tables to reflect shortest path routes (this can be done because network topology does not change) • This will occupy the fewest nodes for each call, but will not necessarily result in routing along the least congested path • Mobile agents • Software agents (worms) move from node to node. Update routing tables based on their observations of the network

  4. Structure of the Paper • Schoonderwoerd et al. Ant-based load balancing in telecommunications networks • Present a new solution – a new kind of distributed mobile agent • Behaviour inspired by that observed in colonies of ants • Evaluate • A simulated network • Measure the rate of dropped calls • Compare with • A different kind of mobile agent • Static routing table

  5. Inspired by Nature • Ants are silly animals that accomplish sophisticated results as a team • Regulating nests temperature within limits of 1˚C • Forming bridges • Raiding particular areas for food • Building and protecting their nest • Cooperating in carrying large items • Finding the shortest routes from the nest to a food source • Mobile agents: we want them to be silly (i.e., simple), but accomplish sophisticated things (load balancing in the communications network)

  6. How Ants Cooperate • Stigmetry – indirect communication through the environment • Produce specific actions in response to local environmental stimuli • These actions in turn affect the environmental stimuli that caused those actions • The new stimuli affect actions of the ants that come to that location • Sematectonicstigmetry • Produce the environmental change: i.e., deposit a ball of mud • Causes other ants to repeat the action, i.e., deposit another ball of mud • Sign-based stigmetry • Deposit pheromones (smelly substances) that cause other ants to behave differently, responding to the presence of pheromones

  7. Example: Laying a Trail (cont.) • Ants lay pheromones as they travel along a trail • A trail’s strength is determined by the amount of pheromones on the trail • Amount of pheromones depends on: • The rate at which pheromones are laid • The amount of pheromones laid – how many ants laid them • How much time has passed since the pheromones were last laid (pheromones evaporate over time) • If many ants follow along the same trail the total amount of pheromones is high – the trail’s strength is high: • Rate of deposit is high • Pheromones laying is recent

  8. Example: Laying a Trail (cont.) Ants started on the right Ants started on the left Shorter path has more pheromones

  9. Potential Problems • Blocking problem • An available route is suddenly blocked • It may take a while to find a new route • Shortcut problem • A better route becomes available • It may take a while to adapt to the new route

  10. ABC: Ant-Based Control • Routing tables are replaced with pheromone tables • Each node in the network has a pheromone table for every other node • Each table has an entry for each neighbour, indicating the probability of using that neighbour as the next hop • Pheromone laying is updating probabilities

  11. Updating Pheromone Tables • At every time step ants can be launched from any node in the network • The destination node is random • Ants move from node to node, selecting the next node according to pheromone tables for their destination node • At each node they update probabilities of the entry corresponding to their source node • They increase the probability associated with the node where they came from

  12. Updating Pheromone Tables (cont.) destination current location 2 source 1 3 4 Update routing table at node 1 for node 3 increase by Δp the probability of taking 4 as next hop

  13. Ageing and Delaying Ants • Recall the system’s objectives: • Find routes that are short; avoid routes that are congested • This is accomplished by ageing and delaying ants • Ageing ants: • Age: the number of time steps the ant has travelled • Δp reduces progressively with the age of the ant • This biases the system to ants who use shorter trails • Delaying ants: • Delay ants at nodes that are congested • Degree of delay correlated with the degree of congestion • This delays updates to pheromone tables leading to congested nodes • Increases age of ants travelling through congested nodes, so their pheromones have a smaller influence on pheromone tables

  14. Routing Calls in ABC Network • Route call to destination D • At the current node, look up the pheromone table for node D • Choose the highest probability in the table • The node corresponding to the largest probability is chosen as the next hop • The call is placed if the route is not congested, otherwise the call is dropped

  15. Solving Blocking And Shortcut Problems • Add a noise factor to ants movement protocol • With probability f ant chooses a random path • This ensures that • Useless routes are used occasionally (so they can be rediscovered if they suddenly become good) • Encourage more rapid discovery of a new route (if it becomes available)

  16. ABC: Putting it All Together • Ants are regularly launched with random destinations on every part of the system • Ants walk according to probabilities in pheromone tables from their destination • Ants update the probabilities in the pheromone table for their source location • They increase the probability of selecting their previous node on the path as the next hop (to their source node) • The increase in probability is a decreasing function of the ant’s age • The ants are delayed on parts of the system that are congested

  17. Other Mobile Agents • Mobile software agent • Load management agent • Parent agent • Travels from node to node • Updates routing table to find the least congested route • Two variations: • Largest minimum capacity (LMC) • Minimum sum of squared utilizations (MSSU)

  18. LMC Total capacity = 10 Spare capacity = 10 - utilization S Node utilization 64 55 Spare capacity 19 55 Minimum capacity of red route: 4 Minimum capacity of blue route: 5 Route with largest minimal capacity: blue 37

  19. LMC Algorithm • Travel from node to node • Label nodes as permanent and temporary • For each node maintain the following fields: • Node ID • Largest minimum capacity of the route from that route to the node’s source agent • The neighbour of the node on this route • Update routing tables to make the node along the LMC route as the next hop • Node along the LMC route is made permanent

  20. LMC Algorithm (Illustration) Link colour indicates the next hop on the way to the S node The algorithm will choose route with largest minimal spare capacity S T P 64 55 P 19 55 P Problem: can result in long routes, occupy many nodes along the way as a result. Does not look at total utilization of the route P 37

  21. MSSU MSSU: Minimal sum of squared utilizations (SU) S Node utilization 6 5 SU = 25 SU = 36 1 5 SU = 25 MSSU of red route: 37 MSSU of blue route: 50 Route with minimal SSU: red SU = 1 3

  22. MSSU Algorithm S Numbers in parenthesis indicate the SSU of the route from the node to S Will make nodes permanent after learning the MSSU of all possible routes Will choose the route with the minimal SSU T(0) T(0) 6 5 P T(25) T(36) 1 5 P T(50) T(37) T(59) 3 P

  23. Network Simulation • A software simulator • Node representation: • A node ID • A capacity – number of simultaneous calls that the node can handle (40) • Routing table with n-1 entries, one for each node. The routing table entry tells us the next hope to take for a given destination node • Probability of being the end node (source or destination of a call) • Spare capacity A B D C Routing table at node C

  24. Network Simulation (cont.) • Calls are generated by a traffic generator • Call parameters: source node, destination node, call duration (170 time steps average) • Call is routed using routing tables, spare capacity of intermediate nodes is reduced • If there is no spare capacity on the route, the call will fail

  25. Experimental Setup • Call probability set: a particular distribution of calls • Adaptation period: run a load balancing mechanism • Test period: measure network performance for the number of dropped calls

  26. Results • What do these numbers indicate? • Which load balancing method performed the best?

  27. Results (cont.) • Percentage of failed calls after stopping load balancing (call probabilities remain unchanged) • What does this tell us about the system?

  28. Results (cont.)

  29. Results (cont.)

  30. Summary • In general ants performed better than other mobile agents • ABC system stores information not only about good current routes, but about good recent alternativeroutes • This allows it to adapt quickly to changes in network conditions • Ants consume less network resources than mobile agents (ants don’t need to store info about all nodes visited) • Ants can work concurrently without affecting each other; only one mobile agent can be active at once • A failure of ant does not hurt the system – other ants will update pheromone tables: the failure of mobile agent affects launching of future agents, so the failure has to be detected

More Related