Simulation

Usage:

After configuring the simulation parameters, the simulation can be run simply as a python script:

python simulation.py

Configuration:

Simulation Parameters are kept in a file called configuration.ini in the working directory and contains the parameters:

[Simulation]
NUM_JOBS = 500
SIMULATION_TIME = 2500.0

[Computer]

[Computer.Cluster]
NUM_SERVERS = 10000

[Computer.Scheduler]
POLICY = LatinSquare

[Computer.Scheduler.LatinSquare]
LATIN_SQUARE_ORDER = 2
PREEMPTION = True
SERVER_SELECTION = RANDOM
NUM_PROBES_PER_BATCH = 1

[Computer.Scheduler.Sparrow]
NUM_SPARROW_PROBES = 1
PREEMPTION = True
LATE_BINDING = True
BATCH_SIZE = 2
SERVER_SELECTION = RANDOM

[Computer.Network]
MEAN_DELAY = 0.0000001
DELAY_STD = 0.0000000001

[Processes]

[Processes.Arrival]
INITIAL_TIME = 0.0
SCALE = 0.000111111111

[Processes.Arrival.Job]
POLICY = Erlang

[Processes.Completion]

[Processes.Completion.Task]
CORRELATED_TASKS = True
HOMOGENEOUS_TASKS = False

[Work]

[Work.Job]
NUM_TASKS = 100

Configure Jobs and Tasks

• A task is a unit of work that is assigned to a server by the scheduler.
• A job is a collection of tasks.

The number of tasks that belong to a job is defined by NUM_TASKS in work.Job.

NUM_TASKS also controls the rate at which jobs will arrive (the shape parameter of the Erlang distribution).

Configure the Computing Cluster

• A server is responsible for the servicing of tasks.
• Each server has a FIFO queue, and services tasks in the order that the scheduler gives them to the server.
• A cluster is a collection of servers.

The number of servers in the cluster is defined by NUM_SERVERS in Computer.Cluster.

There are three currently implemented scheduling policies:

• RoundRobin: Assign tasks round robin to all servers.
• FullRepetition: Assign all tasks in each job to every server in the order that they arrive in. Preempt any remaining tasks in the server queues after the associated job is completed, and move all subsequent tasks up in the server queues.
• LatinSquare: Assign tasks in each job according to the sequence defined by the latin square. Examples for latin squares of order 2, 3, and 6 are given. Higher orders can be constructed as the addition table of the finite group with n elements, where n is the order of the latin square.

The scheduling policy is defined by POLICY in Computer.Scheduler.

Configure the Arrival and Completion Processes:

• The arrival process controls the rate at which jobs arrive.
• The completion process controls the rate at which servers complete tasks.

The arrival process yields interarrival times as an Erlang random variable with a SCALE parameter in Process.Arrivals.

The completion process yields completion times as a standard exponential.

Configure the Simulation:

The simulation is controlled by two parameters:

• NUM_JOBS
• SIMULATION_TIME

NUM_JOBS defines the number of job arrivals to simulate. SIMULATION_TIME defines the maximum simulation time to run for. The simulation will terminate if either:

• The event queue is empty
• the simulated time exceeds SIMULATION_TIME.