Performance Modeling and Benchmarking of Event-Based ... - DVS

60 CHAPTER 4. PERFORMANCE ENGINEERING OF EVENT-BASED SYSTEMS
Depending on the availability of Event tokens in the depository of the Broker, either mode 2-I
(Event available) or mode 2-II (no Event token) is chosen:
1. If an Event token exists the consumer pulls it (transition 3-I ) and disconnects. He will
not reconnect before the Event token is processed.
2. If no Event token exists the consumer disconnects and waits for a specified time interval.
Then, a new Trigger token is generated by transition 3-II and the consumer tries to pull
an Event.
Number of Service Places (Parellel Events) and Issues The pattern offers a simple way
to set the maximum number of Events processed in parallel by defining the initial number of
Trigger tokens.
However, there is a drawback of this approach: Imagine a scenario where we set the number
of parallel events processed by the consumer to two. For the case that no Event token was
available at the broker, two Trigger tokens were transformed to Sleep tokens and moved to
the Timer. After the specified time interval one of the Sleep tokens is processed by the Timer
and transformed back to a Trigger token by transition T3-II : the consumer ’wakes up’ and
establishes a connection to the broker. In the meantime two new Event tokens arrived in the
depository of the Broker. Since there is one Trigger token, only a single Event is moved to the
consumer. The second Event token remains in the depository of the broker until the second
sleep token has been processed by the timer, even if the Consumer has enough resources to
process both Events.
Another approach is presented in Pattern 7, where the consumer pulls as many Events at once
as free resources are available. This avoids opening several connections and allows processing
them as fast as possible.
How to Modify Pattern 6 to Model Thread Pool
By removing the Timer place and transitions T2-II and T3-II the underlying idea of this
approach is made suitable for modeling a thread pool. As illustrated in Figure 4.15, all we need
to implement such a pool are Thread tokens and a Thread Pool ordinary place (corresponding
to Thread tokens, respectively Thread Store).
QPN Definition
Places:
Place Type Description
Producer S Publishes events.
Broker S Stores all incoming events.
Timer Q Timer queue (scheduling strategy: infinite server).
Trigger Store O Stores trigger tokens.
Consumer S Consumes incoming events.
Colors:
Color
Event
Trigger
Sleep
Description
Represents the published event.
Triggers pull commands.
Exists for time between an unsuccessful pull attempt and a reconnect.