1. 12006/9/26 Load Balancing in Dynamic Structured P2P Systems Brighten Godfrey, Karthik Lakshminarayanan, Sonesh Surana, Richard Karp, Ion Stoica INFOCOM 2004. Twenty-third AnnualJoint Conference of the IEEE Computer and Communications Societies, Volume: 4, On page(s): 2253- 2262 vol.4, Publication Date: 7-11 March 2004 Presented by 張肇烜

2. Outline Introduction Introduction Problem formulation and motivation Problem formulation and motivation Background Background Load balancing algorithm Load balancing algorithm Evaluation Evaluation Future work Future work Summary Summary

3. Introduction The last several years have seen the emergence of a class of structured peer- to-peer systems that provide a distributed hash table(DHT) abstraction. The last several years have seen the emergence of a class of structured peer- to-peer systems that provide a distributed hash table(DHT) abstraction.

4. Introduction (cont.) We make the following contributions: We make the following contributions: –We propose an algorithm which to the best of our knowledge is the first to provide dynamic load balancing in heterogeneous, structured P2P systems. –We study the proposed algorithm by using extensive simulations over a wide set of system scenarios and algorithm parameters.

5. Introduction (cont.) Our main results: Our main results: –The algorithm achieves a good load balance. –Our algorithm produces a 99.9th percentile node utilization less than 3% higher than a similar fully centralized load balancer. –Increasing the scalability of the system.

6. Problem formulation and motivation Definitions and goals: Definitions and goals: –Minimize the load imbalance. –Minimize the amount of load moved. Relevance of load balancing: Relevance of load balancing: –Is load balancing of storage feasible? –Why use load balancing for bandwidth?

7. Background Use of virtual servers. Use of virtual servers. –A node may have multiple IDs and therefore owns a set of noncontiguous regions. –Advantages. Static load balancing techniques. Static load balancing techniques. –One-to-one scheme. –One-to-many scheme. –Many-to-many scheme.

8. Load balancing algorithm

9. Load balancing algorithm (cont.)

11. Evaluation Basic effect of load balancing. Basic effect of load balancing.

12. Evaluation (cont.) Load movement vs. 99.9 th percentile node utilization. Load movement vs. 99.9 th percentile node utilization.

13. Future work Prediction of change in load. Prediction of change in load. Balance of multiple resources. Balance of multiple resources.

14. Summary Our simulation results show that our algorithm is effective in achieving load balancing for system utilizations as high as 90% while transferring only about 8% of the load that arrives in the system. Our simulation results show that our algorithm is effective in achieving load balancing for system utilizations as high as 90% while transferring only about 8% of the load that arrives in the system.