cs6235-Eval-feedback-scheduler - CS ‐6235 Real Time...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: CS ‐6235 Real Time Systems Project Proposal Simulation & Evaluation of Feedback based Scheduling Algorithm for Real‐Rate applications on SWiFT Toolkit Archana Asokan (902604654) Subbanarasimhiah B Harish (902641327) Graduate Students, College of Computing Georgia Institute of Technology {archana.asokan,harishutd [email protected] Goal The goal of the project is to use a reconfigurable, dynamic and adaptive toolkit (SWiFT) which provides a feedback control environment to simulate proportion period scheduling algorithm, study its performance for different network based applications using certain benchmarks and compare the results with standard schedulers. Introduction The use of feedback control mechanisms is becoming increasingly important in the scheduling of real time applications in situations of uncertain workload. Real time applications specify the resources and time needed to complete execution. For a long time, the schedulers were purely priority based leading often to conditions of starvation and priority inversion. Also, most schedulers used to be static i.e. once the scheduling is done there is no kind of feedback. To alleviate these problems, a proportion based scheduler with feedback was proposed [1] which dynamically allocates proportion of CPU (Bandwidth allocation in the case of network applications) based on the notion of progress of the application/flow. Motivation The issue of scheduling in Real time scenarios is a very important one because they have a time period within which the task needs to be completed. Feedback control mechanisms seem to be the best solution in terms of monitoring progress and proportion period schedulers overcome the traditional problems of starvation and priority inversion thus motivating us to study and evaluate feedback based scheduling algorithms for network based applications. The idea of feedback and monitoring the task progress using queue/buffer values [2] is what appealed the most to us. To actually see how the scheduler would work by simulating it on SWiFT toolkit and evaluating the scheduler by using application benchmarks seems exciting to us. Design and Architecture The three main components in the feedback based system are:‐ Page 1 CS ‐6235 Real Time Systems Project Proposal ‐ ‐ ‐ The Requirement Monitor The Feedback Controller Packet Dispatcher Figure : Feedback based scheduler Requirement Monitor infers the mismatch between the allocated bandwidth and the bandwidth requirement and defines a Pressure variable based on this. This pressure variable is basically the difference between the flow’s sampled queue fill‐level and the target fill‐level. The Feedback Controller takes in the input from the Requirement monitor. It analysis the pressure values and adjusts the bandwidth of the flow accordingly. The individual progress pressure values are Summed up and passed to a proportional‐integral‐derivative (PID) control to calculate a cumulative pressure. A PID controller combines the magnitude of the summed pressures (P) with the integral (I) and with the first‐derivative (D) of the function described by the summed progress pressures over time. PID control is a commonly applied technique for building controllers to provide error reduction together with acceptable stability and damping. Packet Dispatcher determines the order of the packet transmission through the interface. The packet scheduler in our project is based on Earliest Deadline First (EDF) policy. The dispatcher depends on two basic parameters : proportion and the period allotted by the Controller. Resources ‐ SWiFT :‐ SWiFT is a toolkit for building adaptive system software using a control‐theoretic approach. It allows implementation of feedback control mechanisms within limited domains (network applications in our case) for dynamically reconfiguring them. This composition allows the application to adapt efficiently across a wide range of conditions. Basically the controller helps Page 2 CS ‐6235 Real Time Systems Project Proposal system maintain a reference value for a controlled variable and the control is integrated with the system through monitors and actuators. Benchmarks for evaluation ‐ NetPerf :‐Sophisticated network and file system benchmark. It can be used to measure the performance of many types of networking. It provides tests for both unidirectional throughput and end to end latency. ‐ SSLPerf :‐Open source bench mark to measure the performance of SSL operations. ‐ Hackbench :‐ The hackbench test is a benchmark that uses client and server processes grouped to send and receive data in order to simulate the connections established for a chat room. Each client sends a message to each server in the group. The test creates a group of processes in multiples set by the user. Each client/server pair listens on a socket; the writer writes 100 messages to each socket and the receiver listens on the socket. Therefore in this case, the total numbers of messages sent are 100 X the number of processes specified. ‐ MCS VoIP :‐ measures the quality and performance of Internet connections for Voice over IP (VoIP) usage by simulating real VoIP sessions. Plan of action Page 3 CS ‐6235 Real Time Systems Project Proposal Deliverables PROJECT REPORT ‐ ‐ Evaluation results of the priority period scheduler simulation on the SWiFT toolkit. Comparison of these results with traditional schedulers like EDF. Future work The proportion period scheduler is simulated was at a user level on the SWiFT toolkit. This can be directly built into the Linux kernel so as to avoid latencies which occur at user level. References [1] David C. Steere, Ashvin Goel, Joshua Gruenberg, Dylan McNamee, Calton Pu, Jonathan Walpole, “A Feedback‐driven Proportion Allocator for Real‐Rate Scheduling”. [2] Kang Li, Jonathan Walpole, Dylan McNamee, Calton Pu and David C. Steere ,“A Rate‐Matching Packet Scheduler for Real‐Rate Applications”. [3] Chenyang Lu, John A Stankovic, Gang Tao, Sang H Son, “Design and Evaluation of a feedback control EDF scheduling algorithm” [4] Lui Sha, Xue Liu, Ying Lu, Tarek Abdelzaher, “Queueing Model Based Network Server Performance Control” [5] Ashvin Goel Jonathan, Walpole, Molly Shor , “Real Rate Scheduling” Page 4 ...
View Full Document

This note was uploaded on 10/10/2010 for the course CS 6235 taught by Professor Staff during the Fall '08 term at Georgia Tech.

Ask a homework question - tutors are online