diff options
Diffstat (limited to 'xml/htdocs/main/en/performance.xml')
-rw-r--r-- | xml/htdocs/main/en/performance.xml | 206 |
1 files changed, 0 insertions, 206 deletions
diff --git a/xml/htdocs/main/en/performance.xml b/xml/htdocs/main/en/performance.xml deleted file mode 100644 index cb097558fb..0000000000 --- a/xml/htdocs/main/en/performance.xml +++ /dev/null @@ -1,206 +0,0 @@ -<?xml version='1.0' encoding="UTF-8"?> - -<mainpage id="performance"> -<title>Gentoo Linux Performance Metrics</title> -<author title="Author"><mail link="drobbins@gentoo.org">Daniel Robbins</mail></author> -<version>1.2</version> -<date>8 Oct 2003</date> - -<chapter> -<section> -<title>Gentoo Linux... performance?</title> -<body> -<p>People have often wondered what kind of performance improvement (if any) Gentoo Linux provides over -other Linux distributions. This section is intended to give you an idea of the various factors -that can affect application performance. -</p> -</body> -</section> -<section> -<title>HispaLinux 2003: Gentoo Linux 1.4 vs. Mandrake 9.1 load-times</title> -<body> -<p>On 23 Sep 2003, <b>Jose Alberto Suarez Lopez</b> gave a presentation at <uri -link="http://congreso.hispalinux.es/edicion/2003/presentacion/">HispaLinux -2003</uri> where he demonstrated the load-time performance of the official -<i>Gentoo Linux 1.4</i> release. To compare load-time performance, he used <i>Mandrake -9.1</i>.</p> - -<note>Jose originally thought that Mandrake 9.1 used prelink by default. This -is not the case, and corrections have been made to this document to reflect this.</note> - -<p>These benchmarks compare Mandrake 9.1 against Gentoo Linux 1.4 in both a -un-prelinked (default) and pre-linked state. For the "Gentoo with prelink" test, prelinking was performed as per -the steps of the <uri -link="http://www.gentoo.org/doc/en/prelink-howto.xml">Gentoo Prelinking -Guide</uri> and only the applications being tested were pre-linked. -</p> - -<p>Here is an overview of the operating systems benchmarked for the load-time test: -<ul> -<li><b>Mandrake 9.1</b> (<b>no prelink</b>, default kernel)</li> -<li><b>Gentoo Linux 1.4 for Pentium III</b> (official prebuilt "GRP" packages from CD, <b>no prelink</b>, gaming-sources-2.4.20-r3)</li> -<li><b>Gentoo Linux 1.4 for Pentium III</b> (official prebuilt "GRP" packages from CD, <b>with prelink</b>, gaming-sources-2.4.20-r3)</li> -</ul> -</p> - -<p>The Gentoo Linux 1.4 for Pentium III release uses the following packages: -<ul> -<li>binutils-2.14.90.0.6-r2</li> -<li>gcc-3.2.3-r1</li> -<li>glibc-2.3.2-r1</li> -<li>compilation flags: -O3 -march=pentium3 -fprefetch-loop-arrays -funroll-loops</li> -</ul> -</p> - -<p>Here is an overview of the system used for the load-time test: -<ul> -<li>Pentium III Coppermine 733Mhz</li> -<li>128MB RAM</li> -<li>1461 BogoMIPS (not an authoritative performance measure but people seem to like to know this :)</li> -</ul> -</p> - -<p>To perform the test, three applications (<b>Mozilla 1.4x</b>, <b>NetBeans -3.5x</b> and <b>Kmail</b> (Gentoo version was a newer 3.1.3 version that came -with the release) were repeatedly loaded. Initial load speed as well as -subsequent load speed was recorded.</p> </body> </section> - -<section> -<title>HispaLinux 2003: Load time results</title> -<body> -<note>In these tests, the first column lists initial load time. The second and third columns -list the time required to load the application the second and third time respectively.</note> - -<p><b>Test 1: Mandrake 9.1 without prelink</b> - -<table> -<tr><th></th><th>1st run</th><th>2nd run</th><th>3rd run</th></tr> -<tr><th>Mozilla</th><ti>5.34s</ti><ti>5.45s</ti><ti>5.32s</ti></tr> -<tr><th>NetBeans</th><ti>38.37s</ti><ti>14.58s</ti><ti>20.43s</ti></tr> -<tr><th>Kmail</th><ti>0.42s</ti><ti>0.42s</ti><ti>0.46s</ti></tr> -</table> -</p> - -<p><b>Test 2: Gentoo Linux 1.4 for Pentium III without prelink</b> - -<table> -<tr><th></th><th>1st run</th><th>2nd run</th><th>3rd run</th></tr> -<tr><th>Mozilla</th><ti>2.88s</ti><ti>2.82s</ti><ti>2.88s</ti></tr> -<tr><th>NetBeans</th><ti>18.98s</ti><ti>18.35s</ti><ti>18.51s</ti></tr> -<tr><th>Kmail</th><ti>0.39s</ti><ti>0.97s</ti><ti>0.33s</ti></tr> -</table> -</p> - -<p><b>Test 3: Gentoo Linux 1.4 for Pentium III with prelink</b> - -<table> -<tr><th></th><th>1st run</th><th>2nd run</th><th>3rd run</th></tr> -<tr><th>Mozilla</th><ti><b>2.8s</b></ti><ti><b>2.78s</b></ti><ti><b>2.82s</b></ti></tr> -<tr><th>NetBeans</th><ti>n/a</ti><ti>n/a</ti><ti>n/a</ti></tr> -<tr><th>Kmail</th><ti><b>0.03s</b></ti><ti><b>0.02s</b></ti><ti><b>0.01s</b></ti></tr> -</table> - -<note>The NetBeans test was not performed for Test 3 since NetBeans itself is -not prelinkable. Also, <i>only the tested applications (rather than the entire -system) were prelinked.</i> Prelinking the entire system shouldn't negatively -impact these performance numbers, and could theoretically improve them if external -binaries are loaded.</note> - -</p> -</body> -</section> -<section> -<title>HispaLinux 2003: Load-time conclusion</title> -<body> -<p>Here are the conclusions of the load-time test:</p> -<ul> -<li><b>Mozilla loads nearly twice as fast under Gentoo Linux 1.4 for Pentium III</b> (both initial and subsequent loads)</li> -<li>Initial load time for NetBeans under Gentoo is over twice as fast as on Mandrake 9.1</li> -<li>Subsequent load time for NetBeans seems about even for both Gentoo and Mandrake</li> -<li><b>With prelink, Kmail loads over 10 times faster than Gentoo Linux without prelink or Mandrake 9.1 without prelink</b></li> -</ul> - -<p>To be completely fair, here are some things to understand about this test:</p> -<ul> -<li>Can you conclude that "Gentoo is faster than Mandrake?" <b>No.</b> This is a limited test. It is likely that -Mandrake is faster for some things. <b>Also, we tested load-time performance only.</b></li> -<li>Can you conclude that "Mandrake is faster than Gentoo?" <b>No.</b> Since we have been able to demonstrate -that there are at least a few things that Gentoo does faster. <b>Again, only load-time performance was -tested, and doesn't necessarily correlate with application runtime performance.</b></li> -<li>"I installed distro X and it is a lot slower/faster than expected!" This is very possible, and can -be due to driver and kernel differences. Often, preemptive and low-latency patches reduce CPU performance -while helping interactive response. Sometimes, certain kernels behave poorly on certain hardware. -You may be loading many more things into RAM with your new install than with the old install, or the -new install's applications may eat up more RAM.</li> -</ul> - -<p><b>What can be gleaned from these results?</b> For one, the default -optimizations of Gentoo Linux 1.4 for Pentium III appear to make a -significant difference in "real world" application load-time performance with -Mozilla. The initial load-time speed delta could be explained by Mandrake -having more services loaded, thus requiring some paging out to virtual memory. -However, second and third load-time speed deltas also show a performance -improvement. This points to lowered CPU usage. To confirm, we would need -to get hard numbers of <path>mozilla-bin</path>'s CPU usage during load time (rather than load time itself) from each system, -do a detailed comparison of libraries and compiler toolchains used, as well as optimization settings. -Binary applications like NetBeans cannot benefit as much from any optimizations -used on the system, -although initial load-time for binary applications does seem to be -a lot better under Gentoo Linux. It is not clear why. Possibly kernel differences -or possibly because more free RAM is initially available on Gentoo as compared -to Mandrake. This latter explanation is likely due to the fact that the 2nd and 3rd -test numbers are more or less the same.</p> - -<p>In addition, KDE C++ applications like Kmail seem to show a huge load-time -peformance improvement when prelinked -- <i>over 10 times faster than without prelink</i>. Prelinking seems to be very -beneficial for improving load-time performance for KDE apps, although Mozilla -shows negligible improvement. Could this be due to Mozilla's internal -architecture which prevents it from being able to benefit significantly from -being pre-linked? Or did our tester mistakenly try to prelink -<path>/usr/bin/mozilla</path> instead of of -<path>/usr/lib/mozilla/mozilla-bin</path>? Only time (and another load-time -performance test) will tell! :)</p> - -<p><b>Gentoo Linux tries to deliver the best performance possible in -several ways.</b> First and most important, we -try our best to offer the latest and best free software technologies to our -users, including the latest compiler toolchains. Newer compilers produce faster -code, so by getting these tools into your hands sooner, we can often deliver -better performance. It's important to understand that there is no magic -involved except getting the latest free software improvements to you in a -timely manner.</p> - -<p><b>Here's the second way we attempt to maximize the performance of Gentoo.</b> <i>The official -Gentoo Linux 1.4 release ships with pre-built binary packages</i> that have been -optimized for specific CPU models. We've chosen optimizations that perform well -and are very stable.</p> - -<p> -For example, <uri link="http://store.gentoo.org/index.php?item=16&action=viewitem">Gentoo Linux for -Pentium III</uri> was used in this test, which runs very fast on Pentium III and higher -processors, but will not run on Pentium II or "lower" processors. It takes full advantage -of the Pentium III's instruction set, including MMX and SSE instructions.</p> - -<p> -<b>By providing -tailored builds for each CPU model, you are able to use the full instruction set -of your CPU and receive the best possible system performance.</b> For x86 -systems, we currently offer 1.4 builds for <uri link="http://store.gentoo.org/index.php?item=13&action=viewitem">x86</uri> (generic PC), <uri link="http://store.gentoo.org/index.php?item=14&action=viewitem">i686</uri> (Pentium -Pro and higher), <uri link="http://store.gentoo.org/index.php?item=16&action=viewitem">Pentium III</uri>, <uri link="http://store.gentoo.org/index.php?item=17&action=viewitem">Pentium 4</uri> and <uri link="http://store.gentoo.org/index.php?item=18&action=viewitem">Athlon XP</uri> systems. In addition, -we provide an easy way for users who wish to "build Gentoo from scratch" so that they can optimize and configure Gentoo Linux -as needed. -</p> - -<p>The third way Gentoo can offer improved performance is that the default install -is "opt in" rather than "opt out." That is, your system will only have the applications -installed that you actually need. This can lower RAM usage which can result in lower -CPU and VM usage and an overall faster system.</p> - -<p>If you'd like to purchase Gentoo Linux, please visit <uri -link="http://store.gentoo.org">The Gentoo Store</uri>. Gentoo can also be downloaded -from our various <uri link="http://www.gentoo.org/main/en/mirrors.xml">mirrors</uri>.</p> - -</body> -</section> -</chapter> -</mainpage> |