- Load Balancing
- How CPU shares the task
- Different CPU and multi-tasking application, executing together.
- The algorithm of load balancing:
- Divides the computational load to run it with multiple CPUs. When load balancing is done, it will send a message through
- MPI
- How the load is divided in load balancing?
- The important parameters which are using in load balancing
- CPU Scheduling:
- When there are jobs, it schedules.
- Difference between CPU Scheduling and Load Scheduling
- Load balancer - can be a physical device and it can be also based on the software. Decides how to balance the load.
- F5 Load Balancer - Most expensive load balancer
- Which job is scheduled in the process next in multi programming environment?
- Techniques:
- FCFS
- Priority Based
- Round Robin Based
- Multi Level Feedback Queue
- Load Balancing:
- Assigns to the task to the CPU which is not over loaded.
- Job Scheduling
- Every processor has its own queue for job
- Streams of jobs arriving into the Job Scheduling.
- Local queue: for the jobs that are locally done.
- Every process has its own queue.
- Policies:
- Random: Schedules job according to the probability, it is a simple and state policy.
- Cycle: It stores the information of the previous allocated site and it is semi state policy. It will decide on the runtime
- so if response is not received, it will assign to the other.
- Join The Shortest Queue:
- Stores the size of the local queue in each site, so it will allocate to the queue which has the shortest queue.
- Parameters of Interest:
- Mean Response
- Mean Delay
- Throughput of the system
- Sender Intiated Load Balancing:
- The node that has the higher load will start this process. Decided by sender
- Different policies of transfer, whether the task needs to be transferred to the different site or different location.
- Threshold value is greater, then it will transfer to other node, otherwise if not greater then it is kept in the same node
- Threshold is actually the number of tasks in queue.
- Shortest Location Policy
- Finds the shortest queue, and then depending on how short it is, it will make the decision to select shortest one
- Receiver Initiated Load Balancing:
- Receiver will try to get the task from the loaded site.
- The reason why it will pool again is that the receiver will raise the flag that it is free so that the task can be received.
- Drawback of receiver initiated algorithm
- Most tasks are preemptive so what happens is that the task is stopped when it wants to perform that action so this is the
- drawback as it causes an interrupt.
Recent Pastes