Triple Modular Redundancy (TMR) is a fault-tolerant technique, and it is a form of N-MR. In this method, computations are being done in three simultaneous flows, and in the end, a voter checks the results and chooses the most frequent one between the three outcomes (Fig.1). This scheme will help to cover unwanted faults. There are more to read on: Triple modular redundancy - Wikipedia
The purpose of this program is applying the TMR technique on conventional matrix multiplication. Three threads do matrix multiplication, and a voter chooses the majority as the best result. The code will run on both Windows and Linux systems, but the actual behavior may vary.
Fig1: TMR Diagram
The code is split using predefined compiler macros in two parts, Linux and Windows. Windows version implemented using C++ standard thread library. The major problem with the thread library is that it is not possible to pin an individual thread to a specific CPU core. However, in Linux implementation, the core library is POSIX threads (pthread) that there is a possibility to define the desired affinity mask to assign a thread to a particular core of CPU. Read more on pthread: POSIX Threads - Wikipedia
On a Linux machine, the command g++ -O3 -pthread -std=c++11 Source.cpp -o TMP compiles and ./TMR runs the code. Following running the program, a menu will come up with the following options.
- Exit
- Hardware Concurrency Info. of Machine
- Input Matrices
- Generate Random Matrices with Pre-Defined Maximum Dimension (10x10)
- Show First Matrix
- Show Second Matrix
- Matrix Multiplication
- Show Multiplication Result
- Fault Simulation
- Dump Matrices to File
Putting self-explanatory options aside, number 1 shows maximum threads available on the machine using the standard C++ thread library. Option 8, simulates situations that each thread may face a fault. In theory, the chance of encountering a fault is nearly zero. Therefore, adding some intentional faults is helpful to observe the functionality of this technique.
 Fig.2: Fault Simulation |
 Fig.3: General Running |
|---|
This program was a part of a take-home midterm exam of "Fault-Tolerant Systems Design" at Sharif University of Technology.