Part 7 - Switch Design

Switch Design

Tutorial Goals

The following is a summary of the concepts you will learn in this session.

1. Define a flow diagram.
2. Setup the Routing between Systemic components.
3. Mapping from behavior to hardware platform.

Target

The purpose of this session is to build a performance model that calculates the statistics and identifies system bottlenecks and evaluates the performance of different applications on the target architecture. The block diagram of this system is shown in Figure 1. The system has two identical cores on a shared bus with a cache and SDRAM.

Figure 1: Block Diagram of the Switching System

VisualSim Tutorial Model Location

Open this model in VisualSim from the following location:

File->Open File->$VS/doc/Training_Material/Tutorial/WebHelp/Tutorial/Performance_Modeling/Switch_Tutorial.xml  

Model Objectives

1. Define the sequence of events.
2. Dynamically map each transaction to one of the processing units.
3. Create the routing table and a Central switch that connects all the devices together.
4. Vary the workload speed and evaluate the processing unit throughput and switch buffering.

The analysis determines the system-level impacts of varying the parameters and the system workloads.  The generated reports must capture the latency over time, the utilization and buffering at each device and the effective throughput.   

Blocks Used

System Description

The following information describes the system:

• Block Diagram in Figure 1.
• Supports n number of Processing Devices.
• The flow diagram is shown in Figure 2.

Figure 2: Flow of Packets from Source to Sink after processing in multiple devices

• Parameters and Setup

• Workload Mean rate= 1.0 seconds
• List of Fields required for storing simulation information
• List of Types and their associated processing sequence
• Queue connection for each Device
• Processing delays at the Devices and Switch

Translation to Model Speak

The key concept in modeling is to model all the elements that are relevant to the analysis and abstract out the rest. The first are the list of processing devices that are modeled using SystemResource.  The switch is modeled as a Server. The rest are a series of logic to define the movement of the data and capture statistics.  The model consists of 5 parts - workload, data structure setup, lookup, queue + processing servers with scheduling algorithm, and statistics gathering.  

Model Block Diagram

The information from the system descriptions and assumptions are compiled into this model diagram.

Figure 3: Execute the model

Building the Model

The following are the translation between the flow diagram and the model elements:

• ModelSetup > Digital
• Parameters - Create parameters as follows:
• Num_Switch                = 4  The number of queues in the switch. It shows the number of connected devices

• For Traffic use "Traffic" block. The tutorial shows traffic being generated at the rate of one packet/ second.  

• For setup, use the ExpressionList block.  The ExpressionList block has multiple purposes.  This includes computing statistics, updating information in memories, and modifying/updating/creating field values.
• What is the purpose of the Task_Number field of the Mapper? Try out the Priority for both the Server and the SystemResources.  Notice how the performance changes.  This is because the queues are reordered every time a new packet arrives.

• Use the Mapper block in the flows to connect the behavior with the architecture.
• Use the ExpressionList block to route the packet through the model flow.
• Statistics

• Execute the simulation and correct any errors.

Analysis