SRMS Workbench	SRMS Paper	RTCC Group

Introduction

The SRMS Workbench is a software system developed to demonstrate the notion of Statistical QoS employed in SRMS [AtlasBestavros:1998]. The SRMS Workbench includes: (1) the SRMS schedulability analyzer (QoS negotiator), and (2) a SRMS simulator (Basic SRMS + all extensions). These two components are packaged into a Java Applet that can be executed remotely on any Java-capable Internet browser. For comparison, other scheduling algorithms, including RMS [LiuLayland:1973] and SSJAC [AtlasBestavros:1998] are included.

Through a simple GUI, the SRMS Workbench allows users to specify a set of periodic tasks, each with (a) its own period, (b) the distributional characteristics of its periodic resource requirements (e.g. Poisson, Pareto, Normal, Exponential, Gamma, etc.), (c) its desired QoS as a lower bound on the percentage of deadlines to be met, and (d) a criticality/importance index indicating the value of the task (relative to other tasks in the task set). Once the task set is specified, the SRMS Workbench allows the user to check for schedulability under SRMS. If the task set is schedulable, the SRMS Workbench generates the appropriate allowance for each task and allows the user to create an animated simulation of the task system, which can be executed and profiled. If the task set is not schedulable, the SRMS Workbench informs the user of that fact and suggests (as part of the QoS negotiation) an alternative set of feasible QoS requirements that reflects the specified criticality/importance index of the tasks in the task set.

The SRMS Workbench is available on the Web at http://www.cs.bu.edu/groups/realtime/SRMSworkbench

QoS Specification and Negotiation

To use the SRMS Workbench for QoS Specification and Negotiation, follow the following steps:

• Select the scheduling algorithm with which you would like to experiment.
• Select the number of tasks in the task system with which you would like to experiment. The applet supports 1 to 9 tasks (due to display restrictions).
• Enter the task parameters for each task in the task set. Depending upon the scheduling algorithm chosen, each task may require different parameters.
• Once the tasks are properly entered, check the system schedulability. There are two possible outcomes.
  • If the task set is schedulable, you are all set.
  • If the task set is not schedulable, the system displays a recommended QoS (i.e. that is schedulable). Such recommendations take into consideration the QoS and criticality you had specified in the previous step. You may use this recommended QoS as is (and thus be guaranteed the schedulability of the task set), or else you may use it to guide the choices you make for the next round of schedulability testing. In either cases, you will need to go back to the previous step to modify the task parameters (unless you would like to experiment with the system even though the schedulability test failed).
• Regardless of whether the system is schedulable or not, you will be given the option of creating a simulator for the specified task system to illustrate the behavior of the system under the selected scheduling algorithm.

Task Set Simulation

To start the simulation of a task set that you specified as described above, click on the "Create Simulation" button. As a result, a simulation window (that displays how a resource is scheduled given the specified task system and scheduling algorithm) is created. The top line represents the resource utilization. The lower lines show the time that each task gets on the resource. The highest priority/shortest period task is highest on the screen. Statistics on the deadlines met and missed are gathered on the right of the scheduling display. The quality of service (QoS) is the percent of jobs which make their deadlines. In the display, a flag is displayed on a task's timeline when a new job becomes ready; if the flag is colored, then the job was admitted to the system. Above the flag is a square . The square is black if the previous job met its deadline, and is white if the previous job missed its deadline.

Task Parameter Requirements

Period

An integer greater than 0. A new job is ready at the beginning of every period and is due at the end of every period.

Distribution

The probability density function which describes the execution requirement of the task. Default assumption in RMS is a fixed execution time.

Mean

The average execution requirement of the task. This is required for all distributions. A positive real number is required.

Standard Deviation

A positive real number is required. This further defines those distributions which need a second parameter.

Superperiod

A term used in Statistical Rate Monotonic Scheduling. A task's superperiod is the period of the task with the next longest period in the system. If the task has the longest period, then its superperiod is assumed to be five times its period (arbitrarily).

Budget

Rate Monotonic Scheduling

unused

Statistical Rate Monotonic Scheduling

The task's time budget. The budget is an integer between 0 and the period of the next longer task. The budget is the time available to that task for each superperiod . Either the budget or the QoS can be specified.

Slack Stealing Job Admission Control

A threshold for execution times. Any job with an execution time less than or equal to the budget is automatically admitted to the system and guaranteed. A job with a longer execution time must attempt to admit its excess time requirement. An integer between 0 and the task's period must be entered.

QoS

The Quality of Service is only used by Statistical Rate Monotonic Scheduling. The user can specify either the budget or the QoS for each task. Whichever is not specified will be calculated when the system schedulability is checked. The value entered must be a real number between 0 and 100.

Criticality

Represents how important a task is. An integer larger than 0 must be entered. This value is only used by Statistical Rate Monotonic Scheduling to distribute unpromised utilization to tasks.

Buttons

Check System Schedulability

The schedulability of the system is reported, and the Create Simulation button is enabled. If the scheduler is RMS, then the system is tested for complete schedulability and for average schedulability. If the scheduler is SRMS, then a schedulable system means that the time budgets are guaranteed to each task. If the scheduler is SSJAC, then a schedulable system means that any jobs with execution times lower than their budgets are guaranteed to meet their deadlines. Note: For SRMS, this will also calculate the QoS and/or budget for each task. These calculations are done based upon an array of randomly generated execution times. If a task's superperiod is larger than four times its period, these calculations may take some time.

Create Simulation

This creates a simulation below the task specifications. The simulation uses the previously specified task system and the specified scheduling algorithm.

Start Simulation

This will start the scheduling simulation.

Stop Simulation

This will stop the current simulation and reset to the start of the simulation, so the simulation is just as if it had just been created.

Pause/Restart Simulation

This will pause or restart the simulation, without changing its state.

Step Simulation

This will step the simulation half a time unit.

Remove Simulation

Removes the simulation window from the applet.

Enable/Disable Dual Priority

This exists only with the SRMS algorithm. If dual priority is enabled, then rejected jobs are permitted to run if the resource would otherwise be idle.

Enable/Disable Time Inheritance

This exists only with the SRMS algorithm. If time inheritance is enabled, then, at the end of its superperiod, a task can pass its unspent budget to the task with the next longest period. This preserves guaranteed high priority time as long as possible.

Enable/Disable Overlap Heuristics

This exists only with the SRMS algorithm and is only useful with non-harmonic periods. If overlap heuristics are enabled, then when a job is released in one superperiod and has a deadline in the next, an effort is made to schedule the job using the release superperiod's remaining allowance.