📄 00000135.htm
字号:
<HTML><HEAD> <TITLE>BBS水木清华站∶精华区</TITLE></HEAD><BODY><CENTER><H1>BBS水木清华站∶精华区</H1></CENTER>发信人: saka (机器猫), 信区: Linux <BR>标 题: NT与Linux大比拼的真实情况 <BR>发信站: BBS 水木清华站 (Mon Apr 19 07:36:04 1999) WWW-POST <BR> <BR>Mindcraft Reality Check
<BR>
<BR> E.L. Green, Linux Hardware Solutions
<BR>
<BR>I. Executive Summary:
<BR>
<BR>Recently Mindcraft released a study comparing network performance of Windows <BR>NT 4.0 and Red Hat 5.2. The results
<BR>of such performance benchmarks are of great interest to us and to the Linux <BR>community as a whole. Past benchmarks of
<BR>Linux performance vs. its competitors have allowed us to perform many <BR>enhancements and improvements to the
<BR>performance of the Linux kernel and associated programs, such as improved <BR>network performance (a result of
<BR>benchmarks against FreeBSD), improved thread switch performance (a result of <BR>benchmarks against Windows NT),
<BR>and improved performance of the kernel open() call (a result of benchmarks <BR>against Windows NT).
<BR>
<BR>Fair comparisons are welcomed by those of us in the Linux community. <BR>Unfortunately, this Mindcraft comparison
<BR>appears to be fatally flawed by misconfigurations of Linux, and thus is not <BR>useful for comparing the true performance of
<BR>Red Hat 5.2 versus Windows NT.
<BR>
<BR>II. Personnel Problems with Study
<BR>
<BR>Mindcraft has proven themselves in the past to be masters of expertly tuning <BR>Windows NT. See, for example, their
<BR>comparison testing of Windows NT versus Novell Netware 5.0, which showed <BR>Mindcraft's skill at expertly tuning
<BR>Windows NT to its best. In this study, Mindcraft continues to display their <BR>expertise in tuning Windows NT.
<BR>Unfortunately, they also demonstrate their lack of expertise in tuning Linux. <BR>
<BR>
<BR>Tuning Linux, and any Unix in general, requires a certain level of expertise. <BR>This level of expertise cannot be gained
<BR>within the three-day time period during which this study took place. This <BR>expertise can be gained via coursework from
<BR>companies such as Learning Tree International or Eklektix, via concerted <BR>querying of the USENET newsgroups via
<BR>DejaNews over a period of months, and by other such mechanisms over a period <BR>of months. This level of expertise is
<BR>similar to the level of expertise needed to tune Windows NT, which generally <BR>requires MCSE coursework in order to
<BR>fully master.
<BR>
<BR>A useful comparison test between Windows NT and Red Hat Linux 5.2 would <BR>require that a Linux engineer with similar
<BR>expertise in tuning high-end systems be on hand in order to properly tune the <BR>Linux system. When ZD Labs
<BR>benchmarked Samba versus NT, they invited Jeremy Allison of the Samba team to <BR>assist in setting up their Samba
<BR>configuration. A comparison test of properly tuned Linux and Windows NT <BR>systems would be quite constructive.
<BR>
<BR>Unfortunately, from looking at the large number of technical mistakes (listed <BR>below) made while configuring the Linux
<BR>system, it is quite obvious that such Linux expertise was not on hand. This <BR>violation of basic testing protocol means that
<BR>the results of the test are not useful for valid comparison purposes.
<BR>
<BR>III. Technical Problems with Study
<BR>
<BR>A. Hardware Problems with Study
<BR>
<BR>1) RAID controller Problems: The AMI RAID controller used in the study is <BR>known to have difficulties running under
<BR>Linux. Linux Hardware Solutions was approached several months ago by AMI <BR>regarding using their RAID controllers
<BR>on our systems, and after thorough evaluation, we concluded that the drivers <BR>for those controllers were still of
<BR>beta-quality and not appropriate for a production server. AMI themselves to a <BR>certain extent admits this -- the version
<BR>of the AMI RAID driver used by Mindcraft was version 0.92, i.e., it wasn't <BR>even mature enough to be awarded a 1.0
<BR>version number.
<BR>
<BR>A fair test would use a RAID controller equally well supported by both Linux <BR>and Windows NT, such as the ICP-Vortex
<BR>GDT line of RAID controllers used by Linux Hardware Solutions, or the Mylex <BR>line of RAID controllers used by VA
<BR>Research Linux Solutions. In reality, a company buying a server of this size <BR>for use with Linux is going to buy one
<BR>optimized for Linux with an appropriate RAID controller, rather than buy an <BR>off-the-shelf DELL server with a poorly
<BR>supported RAID controller.
<BR>
<BR>2) Disk Configuration Problems with Study: Both Windows NT and Linux were <BR>installed simultaneously on the
<BR>tested system. The drives were partitioned such that Windows NT occupied half <BR>the drive, and Linux occupied the other
<BR>half. Unfortunately, it is not stated within the study whether Linux was on <BR>the inner half or outer half of the drive. Note
<BR>that the outermost tracks of a hard drive have approximately 1.5 to two times <BR>the data transfer rate of the innermost
<BR>tracks, due to the larger number of sectors spinning past the head in a <BR>second's time.
<BR>
<BR>B. Linux Configuration Problems with Study
<BR>
<BR>1) Failure to configure Linux with large memory support A quick search of <BR>DejaNews with the parameters "2
<BR>gigabyte memory linux" swiftly turned up a message on how to configure Linux <BR>to access up to 2 gigabytes of physical
<BR>memory. In their test, Mindcraft had Linux accessing only 96% of the memory <BR>that they had allocated for NT. This
<BR>would be an immediate 4% performance hit in most file serving situations.
<BR>
<BR>2) Failure to tune buffer allocation parameters Like all Unix-type operating <BR>systems, the Linux operating system has
<BR>various tuning parameters to control the allocation of buffer memory, number <BR>of file handles available, and number of
<BR>inode entries allowed. The default Linux 2.2 kernel will use up to 60% of <BR>physical memory as disk buffer cache, will allow
<BR>up to 4096 open files, and will allow up to 12288 inodes. This is optimal for <BR>applications serving, but is completely
<BR>inappropriate for file or web serving on a 1gb machine. Tuning the file <BR>buffer size so that more than 60% of memory can
<BR>be used (90% in this example) can be accomplished by issuing the following <BR>command:
<BR> echo "2 10 90" >/proc/sys/vm/buffermem"
<BR>
<BR>This is documented in the file /usr/src/linux/Documentation/sysctl/vm.txt <BR>along with many other tuning parameters,
<BR>such as the 'bdflush' parameter. Jeremy Allison of the Samba team claims a <BR>factor of two improvement in Samba
<BR>performance when these parameters are set properly.
<BR>
<BR>Please note that queries of the DejaNews archives looking for queries by <BR>MindCraft researchers regarding tuning the
<BR>Linux 2.2 kernel during the relevant time period turned up only one possible <BR>(but not probable) message, which asked a
<BR>question about network card tuning. During the time span of January 1, 1999 <BR>⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -