1 Introduction to Load Balancing: l Definition of Distributed systems. Collection of independent loosely coupled computing resources. l Load Balancing.
Published byModified over 5 years ago
Presentation on theme: "1 Introduction to Load Balancing: l Definition of Distributed systems. Collection of independent loosely coupled computing resources. l Load Balancing."— Presentation transcript:
1 Introduction to Load Balancing: l Definition of Distributed systems. Collection of independent loosely coupled computing resources. l Load Balancing is a “pre-step” to scheduling.
2 §Random Arrival of user tasks. §Varied Computing resources of different hosts. §Homogeneous Vs Heterogeneous systems. Motivations:
3 Basic Issues: §Definition of LOAD and Performance as the basics of Load Balancing. § Criterion for Load Balancing: Process based factors. Resource based factors. Algorithmic factors. § Efficient Evaluation of these criterion.
4 Load Balancing Vs Sharing: §Both attempt to maximize response time. §Balancing implies that load has to equalized rather than just shared. §Hence Load Balancing is a special Case of Load Distribution policy. §Load Balancing has more over head.
6 Centralized Load Balancer: §Advantages: l Single Host Implementation. l Highly adaptive. l No Overhead on individual hosts. l Better State Consistency. l Centralized transfer of tasks. §Disadvantages: l More expensive. l Single point of failure. l Relatively lesser scalable.
7 Load Balancing Architectures: Peer to Peer Architecture:
8 De-Centralized Load balancer: §Advantages: l No Single point of failure. l Independent unit of operation for each host. l Highly scalable. §Disadvantages: l Complex implementation. l Overhead on each host due to load of the algorithm. l Lesser level of adaptability to heterogeneous environments.
9 Peripheral Components: §Client Programs: l No knowledge of location of execution i.e. “Location Transparency”. l Priority, type of process may influence Load Balancing decisions. §Resources: l Decide the granularity of Load Balancer. l Different types of resources.
10 l Status of resources & collection mechanisms. Broadcast.(periodic, demand driven, & state change) Kernel based monitor. (periodic, demand driven, & state change) Publish - Subscribe Model. Peripheral Components:
11 Peripheral Components: §Communication network: l Speed, reliability & adaptability issues. l Hetrogeneous requirements for networks.
12 Load balancing Details: §Load Balancing policies: l Static l Dynamic Adaptive(Learning) Non-Adaptive.
13 Load Balancing details: §Load & Performance Metrics: l Load : Index of CPU Queues, CPU utilization etc. l Performance: Measured as an index of average response time for client processes.
14 Load balancingDetails: §Type of task transfers: l Preemptive: Process Migration required. Generally more over head involved as the entire process state is transferred. l Non- Preemptive: Based on initial task placement. Simpler & more efficient.
15 Load Balancing Details: §Composition of an Load Balancing Algorithm: l Transfer policy: Determines the state of a node I.e sender or receiver. l Selection policy: Determines “which” task would be transferred. l Location policy: “where” to transfer. l Information policy: State maintanence.
16 Load Balancing Details: §Stability of a Load Balancing Module: l Algorithmic stability. l System stability. l In effective Vs effective algorithms. l Stability & Effectiveness of a Load Balancing Algorithm.
17 Conclusion: §Load balancing forms an important strategy for the improvement of the average response time for any user process. §Internet as a major boost for its application. §Load balancing widely used as a major component of clustered solutions. §Advances in technologies relating to peripheral components leads to more focus on efficient implementation of the load Balancing Algorithm.