Using GEM5 as a tool. We want to do comparison between Full System Simulation and Processor Simulator.
M5 or gem5 full simulator:
-> Full simulator which also simulates the OS(possibly unix)
-> Execute same program on this M5 and simplescalar and plot various matrix
-> like l1 cache misses, l2 cache misses, simulation time, branch prediction
-> Compare both the results
Objective 1: The objective here should be to be able to recommend for what type of programs Full simulation is helpful and for what processor simulation is helpful
Objective 2: Is there any difference between the matrix for same program, why are they different, which simulation is better
Objective 3: Is processor simulation always recommended ?
Objective 4: Can we modify Simplescalar to handle some library calls(which might be executing frequentlly) which will improve simulation ?
Constraints : 1. Set same architecture in both simulators 2. Set same cache and other reg size 3. If possible instruction set should be same
- No need to simulate on SimpleScalar, we can use GEM5 for both.
- Try to take measurements of many metrics for a program on different configurations
- Simulate only some billion or million instructions
- Don't simulate the whole program.
- Try to simulate two or three programs only
- clusters should not have more than 2 nodes