CSC/ECE 506 Fall 2007/wiki1 12 dp3: Difference between revisions
| Line 8: | Line 8: | ||
In this introduction section, we describe why performance issue is important for parallel computer architects and what the metrics of performance are. | In this introduction section, we describe why performance issue is important for parallel computer architects and what the metrics of performance are. | ||
As we already know performance is a very important issue in uniprocessor system where architects always try to improve performance of system in terms of execution time by using several techniques such as minimizing memory access time (to access the data fast), designing hardware which can execute many instruction in parallel and possibly faster ( micro level parallelism extraction).Consequently, performance issue plays a big role in parallel computing for several reasons like data is shared among many processors and hence processes on different processors need to communicate efficiently and coherently. | As we already know performance is a very important issue in uniprocessor system where architects always try to improve performance of system in terms of execution time by using several techniques such as minimizing memory access time (to access the data fast), designing hardware which can execute many instruction in parallel and possibly faster ( micro level parallelism extraction).Consequently, performance issue plays a big role in parallel computing for several reasons like data is shared among many processors and hence processes on different processors need to communicate efficiently and coherently. | ||
To make our point more precise, let us consider the following example: | To make our point more precise, let us consider the following example: Assume we want to run a program which takes 100ms on uniprocessor with no pipeline scheme. However, we also know that the full program can be decomposed in many processes and these processes can be run on different machine. So basically, in best case we get the speed up of ‘n’ where is the number of processors available. We might have virtually divided the computing load but these processes are not independent so in order to complete the program they do need to communicate to each other for data sharing, synchronization etc. So there is communication overhead involved in parallel computing. A wise architect would not like to have any parallel system where communication overhead overwhelms speed up achieved by dividing the work load. So we need to analyze the performance considering various trade-offs. | ||
There are three basic performance metrics. | There are three basic performance metrics. | ||
1. Latency : Time taken by a operation to get completed.(seconds per operation) | 1. Latency : Time taken by a operation to get completed.(seconds per operation) | ||
Revision as of 01:12, 6 September 2007
Sections 1.3.3 and 1.3.4: Most changes here are probably related to performance metrics. Cite other models for measuring artifacts such as data-transfer time, overhead, occupancy, and communication cost. Focus on the models that are most useful in practice.
Communication and Replication
Performance
Introduction
In this introduction section, we describe why performance issue is important for parallel computer architects and what the metrics of performance are. As we already know performance is a very important issue in uniprocessor system where architects always try to improve performance of system in terms of execution time by using several techniques such as minimizing memory access time (to access the data fast), designing hardware which can execute many instruction in parallel and possibly faster ( micro level parallelism extraction).Consequently, performance issue plays a big role in parallel computing for several reasons like data is shared among many processors and hence processes on different processors need to communicate efficiently and coherently. To make our point more precise, let us consider the following example: Assume we want to run a program which takes 100ms on uniprocessor with no pipeline scheme. However, we also know that the full program can be decomposed in many processes and these processes can be run on different machine. So basically, in best case we get the speed up of ‘n’ where is the number of processors available. We might have virtually divided the computing load but these processes are not independent so in order to complete the program they do need to communicate to each other for data sharing, synchronization etc. So there is communication overhead involved in parallel computing. A wise architect would not like to have any parallel system where communication overhead overwhelms speed up achieved by dividing the work load. So we need to analyze the performance considering various trade-offs. There are three basic performance metrics. 1. Latency : Time taken by a operation to get completed.(seconds per operation) 2. Bandwidth: The rate at which operations are performed ( operations per second) 3. Cost: It is the measure of impact operations have on total execution time of the program. (latency times number of operations) In parallel computing many operations are performed concurrently so relationship among performance metrics is not simple and we need to consider the performance for communication operations in parallel computing and see how they are measured using these three matrices. As data transfer operations are the most frequent type of communication operation, we will first discuss that.
Artifacts of Measuring Performance
Data Transfer
Overhead and Occupancy
Communication Cost
- Linking to previous Artifacts.
- How to measure- Mathematical Equation
- Quantitative Measure
- Implication of such measurement
- Advantage and Disadvantage
- New Advances
New Artifact
- not decided yet!