Objectives
- Create a benchmark and measure the performance of multiple computers.
- Define criteria for measuring the performance of a computer
- List, in order of performance, all computers evaluated by the class.
Attention: This exercise is individual, with a collective final result. You must take the measurements and place them on the wiki until 07/01. The report can only be made after this date.
Before you start
You should remember the concepts of performance measures studied in MC722. Here are some questions to guide your exercise:
- What is a benchmark?
- How many programs should there be in a benchmark?
- Which measure should be used?
- How to obtain this measure?
- Is the measure used reliable?
- Is the scale used reliable?
- Is there an error in these numbers? How to minimize it?
- What does it mean to say that one computer is X times faster than another?
- Do you know how to differentiate a sequential from a parallel program?
- If your program is sequential, does it make a difference to run it on a parallel machine?
- If your program is parallel, how do you measure your performance on a sequential machine?
You should know how to answer all of these questions, but it is not necessary to put the answers in the report. Do not write the report without knowing the answers to these questions.
Activity
This exercise is the result of collective work, where each component must deliver a report and perform a part of the activity. Each student must perform only one
the following activities:
- Define a program to measure the performance of a computer: Choose a program (it cannot be part of an existing benchmark) that can be used to measure the computer's performance. Activities and recommendations:
- Do not use a program that is, by itself, a benchmark
- The program must have an open source code to allow it to be compiled and run on multiple machines (you don't need to run it on multiple machines)
- Define, if necessary, a set of standard input (s) and output (s) for your program, so that all executions are identical in behavior
- Prepare a script
or define a command line so that the program runs with the correct settings (if any) and parameters
- Be careful not to choose programs that can generate performance numbers that are susceptible to big errors (very small numbers, for example). Run times less than 10s can cause accuracy errors
- Define your performance and measurement criteria on a base computer. It is necessary to measure performance more than once and consolidate the measures
- Publish the information above so that other students in the class can reproduce their data on other computers. Use this spreadsheet
to put your data
- Justify the reason for choosing this program, describing it briefly
- Don't forget to put the computer's specifications in the spreadsheet.
- Run the three programs defined in item 1 on a computer: Choose three of the programs that have been released by your colleagues and run them on another computer that has not been used. Put the performance information on the same page described in item 1. Don't forget to put the computer's specifications on the spreadsheet.
- Run a program defined in item 1 on three different computers: Choose a program that was released by one of your colleagues and run it on three computers other than those that have already been used. Put the performance information on the same page described in item 1. Don't forget to put the computer's specifications on the spreadsheet.
- Fill in spreadsheet spreadsheets: In case of difficulties in finding programs or computers to carry out activities 2 or 3, fill in 3 different cells of the spreadsheet. Place the performance information on the same page described in item 1.
Based on the results available in the spreadsheet, make your report comparing the performance of all computers. Use all points provided. Do not forget to inform, for each measurement, the settings of the computer that was used. Rank computers in order of performance.
Delivery
Send a report of just 1 page describing the activity performed. Use a section / subsection only to describe your individual activity. Analyze and comment on the result.
- Delivery fee: 11/01 at 8am
by email.
- Format: file PDF.
- Email Subject: MC723: E1 Delivery - o_numero_do_your_ra_sem_as_letras_ra