Client-Server Assignment for Internet Distributed Systems Overview • Introduction • Problem Definition • Problem Model • Solution • Conclusion Introduction Internet - Distributed System Example: Email,IMS Features: 1. Communication Load ο Clients assigned to two different servers. ο Clients assigned to same server. 2. Load Balancing ο Use fewer servers. Servers are heavily loaded Observations: Problem Definition ο Optimal client server assignment for a pre-specified trade-off between load balance and communication load. Emerging Applications: 1. Social networks Eg: Facebook 2. Distributed database system, Eg: MapReduce Communication Model Initially assign clients to a system with 2 servers (Sa, Sb) Then we extend the 2-server solution to multiple servers. Xi = 1, client i is assigned to Sa Xi = -1, client i is assigned to Sb ππ,π : data rate from client i to client j. Communication Load ο ππ,π if i and j are assigned to same server. ο 2ππ,π if clients are assigned to 2 different servers. ο ππ,π = 0 Total communication load, πΆπ‘ππ‘ππ = ππ,π + π π π π 1 − π₯π π₯π × ππ,π 2 If i and j are assigned to different servers, π₯π π₯π = -1 Load Balance πΆπ΄ = π,π πΆπ΅ = 1 + π₯π ∗ ππ,π + 4 1−π₯π π,π 4 π,π ∗ ππ,π + Load balance, D = πΆπ΄ − πΆπ΅ 1 + π₯π ∗ ππ,π 4 π,π 1−π₯π 4 ∗ ππ,π D can be expressed as, π·= π,π π₯π + π₯π 2 ∗ ππ,π Refer link Adding D to objective function will make the function non-quadratic. Hence we modify D, πΆπππππππ = (πΆπ΄ − πΆπ΅ )2 Equivalent formula of D, D= π π₯π ∗ π π , where π π = 1 π 2∗ (ππ,π + ππ,π ) Refer link As, π₯π ∗ π₯π = 1, πΆπππππππ = Refer link 2 π π π + π≠π (π₯π π₯π ) ∗ ( π π π π ) Optimization problem: Minimize: ∝ πΆπ‘ππ‘ππ + 1 −∝ πΆπππππππ Subject to : π₯π 2 = 1, π₯π ∈ π. 1 ≤ π ≤ π. Where: πΆπ‘ππ‘ππ = πΆπππππππ = π π ππ,π 2+ π π π + π π 1−π₯π π₯π 2 π≠π (π₯π π₯π ) × ππ,π ∗ ( π π π π ) ∝ is an arbitrary co-efficient (0≤ ∝≤1) Objective function : minimize π≠π π€π,π π₯π π₯π +πΆ Where we define, 1 π€π,π = − ∝ ππ,π + 1 −∝ π π π π 2 3 πΆ = ∝ π≠π ππ,π + 1 −∝ π π π 2 2 Refer link Semidefinite Programming ο Semidefinite programming is a class of convex optimization. ο π π : set of real Symmetric ππ₯π matrices. ο A matrix A ∈ π π is called positive semidefinite if π₯ π π΄π₯ ≥ 0, for all π₯ ∈ π π ο It satisfies strict quadratic programming Solution: minimize: tr(π π π) subject to: π’π,π = 1, 1 ≤ π ≤ n π ππ π ππππππππππ‘π. Solution Matrix = ππ ππ W-> Matrix with diagonal elements 0 and Wi,j U -> symmetric & Positive semidefinite matrix Conclusion 1. Hard problems could be formulated as a optimization problem and solved. 2. optimization problems, are widely used in tremendous number of application areas, such as transportation, production planning, logistics etc. Presented by : Swathi Balakrishna Extra information: Transform program into Vector program: Minimize: π≠π π€π,π π£π . π£π + πΆ Subject to: π£π . π£π = 1, π£π ∈ π π£π ∈ π π , π£π ∈ π Vector programming -> Semidefinite programming W-> Matrix with diagonal elements 0 and Wi,j U -> symmetric & Positive semidefinite matrix minimize: tr(π π π) subject to: π’π,π = 0, 1 ≤ π ≤ n π ππ π ππππππππππ‘π. Solution Matrix = ππ ππ Cholesky Factorization: Obtain V= (π£1 , π£2 , … , π£π ) , Satisfying π π π = ππ ππ . Final solution: Round n vectors (π£1 , π£2 , … , π£π ) to n integers (π₯1 , π₯2 , … , π₯π ).