Adaptive Execution of Scientific Workflow Applications on Clouds - Cloud Computing with e-Science Applications - page 82

Agent-based simulation . In this phase, we simulate the actions and
interactions of autonomous agents. This agent-based simulation con-
sists of agent granularity, adaptive agent process, decision-making
heuristics, and agent interactions. Agent granularity refers to the
number of agents specified at various scales. The adaptive agents pro-
cess refers to the action that an agent takes when a situation occurs
(redefining the decision-making heuristics). Decision-making heuris-
tics refer to rules or behaviors of an agent. Agent interaction refers to
the complexity of communications or interactions between agents.

There are three types of agents in the S 3 application: commuters, stations,
and trains. Each of these agents has its own attributes, adaptive agent pro-
cess, decision-making process, and agent interactions, as summarized in
Table 4.1. The order of complexity in this phase is O( n 3 ) due to the interac-
tions between agents on simulation time interval or O( tn 2 ), where n repre-
sents the number of agents and t represents the simulation time steps.

Data requirements . The size and quantity of the data set that is generated
is large. The size of the data can easily take up a few gigabytes each day.
For example, the data set consists of 7 days of public transportation
journeys for each individual, with approximately 3 million journeys
per day. As for the agent-based simulation, we simulate the growing
population as 6.9 million. This translates into approximately 14 mil-
lion journeys (travel and return) performed for each simulated day.
Computation requirements . For agent-based simulation, millions of
agents are created to simulate the future infrastructure and dynam-
ics of the transportation system in Singapore. In total, the system
manages 7 million agents that have their own attributes, adaptive
agent process, decision-making process, and interactions with other
agents. Furthermore, there is complexity of agent interactions and
tracking for the simulation interval at 1-second granularity.

TABLE 4.1
Agent-Based Simulation Characteristics
Commuter Agents

Station Agents Train Agents

Agent granularity 6.9 million agents 90 agents Approximately 200

agents

Attributes 12 attributes 9 attributes 16 attributes

Adaptive agent
process

1 adaptive process — 2 adaptive processes

Decision-making
heuristics

5 decision-making
heuristics

2 decision-making
heuristics

5 decision-making
heuristics

Agent interactions • Station

• Train

• Commuter
• Train

• Commuter
• Station