Cse 221 Project – Winter 2003 System Performance Measurements on RedHat Linux 0 and Windows xp 1 Project Members
Download 205.68 Kb.
|
- Navigate this page:
- 1 Introduction
- 2 Project Overview
|
CSE 221 Project – Winter 2003 System Performance Measurements on RedHat Linux 8.0 and Windows XP 5.1 Project Members: Urvashi Rao Venkata Reena Mathew
1 Introduction 5 2 Project Overview 5 3 Memory 6 3.1 Cache Vs RAM Vs Swap Disk 6 3.2 Code/Memory Optimizations 7 3.3 Looping and Sub-Looping 8 4 Disk IO 10 4.1 Disk access (read-write-read) 10 4.2 Sequential Vs Random Reads 14 4.3 Read Vs mmap 16 4.4 Remote File Access 19
5.1 System Calls 20 5.2 Function performance with different implementation styles. 21
6.1 Integer and Float Operations 23 7 Linux Vs Windows 24 7.1 Memory 24 7.2 Disk IO 26 7.3 Process 28 8 Conclusion 29 Appendix 1 30 References 31 Table of Figures Figure 1 : Code Segment for Memory Access 6 Figure 2 : Graphs for Memory Access 6 Figure 3 : Graph for Memory Access for Code plus Data 8 Figure 4 : Code Segment for Looping and Sub-Looping 8 Figure 5 : Graphs for Sub-Looping with 8192KB on Linux 9 Figure 6: Graphs for Sub-Looping with 8192KB on Windows 9 Figure 7 : Code Segment for Disk Access using 'fread' and 'fwrite' 10 Figure 8 : Graphs for Disk Read And Write Access on Linux 11 Figure 9 : Graphs for Disk Read And Write Access on Windows 11 Figure 10 : Code Segment for Disk Access using 'open' and 'close' 12 Figure 11 : Graphs for Disk Read And Write Access on Linux 12 Figure 12 : Graphs for Disk Read And Write Access on Windows 13 Figure 13 : Code Segment for Disk Access using 'mmap' and 'munmap' 13 Figure 14 : Graph for Disk Access using 'mmap' and 'munmap' 14 Figure 15 : Code Segment for Sequential and Random Disk Access 14 Figure 16 : Graphs for Sequential and Random Disk Access on Linux and Windows 15 Figure 17 : Graphs for Sequential and Random Disk Access on Linux to determine Page Read Ahead Size 16 Figure 18 : Graph for Sequential and Random Disk Access on Windows to determine Page Read Ahead Size 16 Figure 19 : Code Segment for 'read' Vs 'mmap' 17 Figure 20 : Graphs for Disk Access using 'read' and 'mmap' 17 Figure 21 : Individual graphs for 'mmap' and normal ' read-write' 18 Figure 22 : Table for read and mmap timings 18 Figure 23 : Code Segment for Local and Remote Disk Access 19 Figure 24: Graphs for Local and Remote Disk Access 19 Figure 25 : Code Segment for System Calls 20 Figure 26 : Table of performance of system calls 21 Figure 27 : Code Segment for different call types 22 Figure 28 : Graphs for Inline Vs Function Vs Recursion calls on Linux 22 Figure 29 : Graph for Inline Vs Function Vs Recursion on Windows 23 Figure 30 : Graphs for Forks 23 Figure 31 : Code Segment for Arithmetic Operations 24 Figure 32 : Graphs for Arithmetic Operations 24 Figure 33 : Graphs comparing Linux and Windows memory access bandwidths 25 Figure 34 : Graphs for Code plus data with different levels of optimization on Linux 25 Figure 35 : Graphs for Code plus data with different levels of optimization on Windows 25 Figure 36 : Graphs comparing Linux and Windows with O3 optimization level 26 Figure 37 : Graphs comparing the Sub-Looping between Linux and Windows 26 Figure 38 : Graphs comparing disk read bandwidths for Linux and Windows 27 Figure 39 : Graphs comparing disk write bandwidths for Linux and Windows 27 Figure 40 : Graphs comparing Sequential and Random disk access between Linux and Windows 28 Figure 41 : Graphs for comparing the disk accesses for Page Read Ahead between Linux and Windows 28 Figure 42 : Graphs comparing the times for different call types on Linux and Windows 28 1 IntroductionThis project involves measuring the performance of an Operating System (OS) based on hardware architecture. It discusses a set of metrics that can be used to evaluate the performance of certain features of an Operating System. Measurements were made on two Operating Systems, RedHat Linux 8.0 and Windows XP 5.1. The measurements for each operating system have been evaluated and compared between the two operating systems. Observations have been discussed and conclusions have been made with respect to each type of measurement. 2 Project OverviewThe goal of the project has been to evaluate the performance of two different operating systems on the same hardware. The metrics chosen fall into the categories of memory access, disk access, processes and arithmetic operations. Architecture Hardware – Dell Inspiron 2650 Processor – Intel(R) Pentium(R) 4 CPU 1.70 GHz Cache Size – 512 KB RAM Size – 256 MB Hard Disk – 30 GB Operating Systems (i) Redhat Linux Version 8.0 (ii) Microsoft Windows XP Version 5.1.2600 Service Pack 1.0 - Cygwin 1.3.20-1 (UNIX environment for Windows) Compiler gcc3.2 – C compiler (on Linux and Cygwin) Benchmarking and performance measurements correspond to sampling a multidimensional space spanned by non-orthogonal axes. Measurements are often not completely isolated from the effects of related factors. Wherever possible, we have attempted to design our metrics to minimize this interference, and to emphasize/isolate trends corresponding to the aspect being measured. In other places we have attempted analyses that take into account major factors affecting the quantity being measured. We have measured performance parameters and have tried to verify our results with documented features of the OS (for example – kernel buffer sizes) to see if these OS specifications are actually 'visible' from the application level. More than to get absolute numbers, out tests attempt to highlight the way the OS manipulates the hardware, when different code implementations are used. These observations can be analysed in terms of efficient software design, to use algorithms that use the behaviour of the OS to its advantage, or equivalently, that take measures to avoid common pitfalls. We have done all our tests with compiler optimization (O3). Wherever we observed anomalies, we re-ran the tests without this optimization to check if the anomalies were due to it. Our idea behind this is that we wanted to measure these performance numbers based on the best possible performance (i.e. with optimization). We are aware of the fact that the effects of compiler optimization will be different for our test code and any application that may be designed keeping our results in mind, and we have tried to reduce this effect by measuring relative performance. We made absolute measurements only for computations like memory / disk accesses and atomic arithmetic operations which were performed with compiler optimization to get figures for best possible performance. For all tests, the same code was used on Linux and Windows. The “Cygwin” interface was used to translate the Unix kernel calls into equivalent (nearly) functionality on Windows. This does not guarantee an exact functional emulation of the Linux test codes, but is the closest we could get. Note: In the rest of the report, the terms ‘Linux’ and ‘Windows’ correspond to the specifications mentioned above. Download 205.68 Kb. Share with your friends:
|