00000135.htm

一份很好的linux入门资料

through April 12, 1999, there were no
&nbsp;<BR>
messages&nbsp;posted&nbsp;to&nbsp;the&nbsp;Usenet&nbsp;asking&nbsp;about&nbsp;memory&nbsp;or&nbsp;buffer&nbsp;tuning&nbsp;for&nbsp;large&nbsp;&nbsp;<BR>
memory&nbsp;machines.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
3)&nbsp;Use&nbsp;of&nbsp;inappropriate&nbsp;kernel&nbsp;version&nbsp;The&nbsp;Linux&nbsp;2.2.2&nbsp;kernel's&nbsp;network&nbsp;stack&nbsp;&nbsp;<BR>
had&nbsp;severe&nbsp;performance&nbsp;problems
&nbsp;<BR>
interacting&nbsp;with&nbsp;Microsoft&nbsp;clients.&nbsp;For&nbsp;more&nbsp;information,&nbsp;see&nbsp;the&nbsp;Kernel&nbsp;&nbsp;<BR>
Traffic&nbsp;newsletter&nbsp;for&nbsp;March&nbsp;4,&nbsp;1999.&nbsp;Given
&nbsp;<BR>
the&nbsp;known&nbsp;problems&nbsp;with&nbsp;the&nbsp;2.2.2&nbsp;kernel,&nbsp;either&nbsp;the&nbsp;2.2.1&nbsp;kernel&nbsp;should&nbsp;have&nbsp;&nbsp;<BR>
been&nbsp;used,&nbsp;or&nbsp;the&nbsp;2.2.3&nbsp;kernel&nbsp;should&nbsp;have
&nbsp;<BR>
been&nbsp;used.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
4)&nbsp;Allocation&nbsp;of&nbsp;Swap&nbsp;not&nbsp;Specified&nbsp;Linux&nbsp;requires&nbsp;a&nbsp;swap&nbsp;partition&nbsp;for&nbsp;best&nbsp;&nbsp;<BR>
operation.&nbsp;If&nbsp;the&nbsp;swap&nbsp;space&nbsp;is&nbsp;allocated&nbsp;on
&nbsp;<BR>
multiple&nbsp;drives&nbsp;(one&nbsp;swap&nbsp;partition&nbsp;per&nbsp;drive),&nbsp;the&nbsp;Linux&nbsp;VM&nbsp;system&nbsp;will&nbsp;&nbsp;<BR>
utilize&nbsp;the&nbsp;drives&nbsp;in&nbsp;parallel,&nbsp;effectively&nbsp;using
&nbsp;<BR>
them&nbsp;as&nbsp;a&nbsp;RAID0&nbsp;(striping)&nbsp;array.&nbsp;Unfortunately,&nbsp;we&nbsp;don't&nbsp;know&nbsp;how&nbsp;swap&nbsp;space&nbsp;&nbsp;<BR>
was&nbsp;allocated&nbsp;in&nbsp;the&nbsp;Mindcraft&nbsp;test,&nbsp;and
&nbsp;<BR>
thus&nbsp;cannot&nbsp;duplicate&nbsp;their&nbsp;configuration.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
C.&nbsp;Samba&nbsp;Configuration&nbsp;Problems&nbsp;
&nbsp;<BR>

&nbsp;<BR>
1)&nbsp;Recompilation&nbsp;of&nbsp;Samba&nbsp;with&nbsp;inappropriate&nbsp;settings:&nbsp;Red&nbsp;Hat&nbsp;Software&nbsp;has&nbsp;&nbsp;<BR>
the&nbsp;Samba&nbsp;2.0.x&nbsp;server&nbsp;available
&nbsp;<BR>
precompiled&nbsp;on&nbsp;their&nbsp;web&nbsp;site&nbsp;as&nbsp;a&nbsp;pre-packaged&nbsp;upgrade&nbsp;to&nbsp;go&nbsp;along&nbsp;with&nbsp;the&nbsp;&nbsp;<BR>
2.2&nbsp;kernel.&nbsp;For&nbsp;some&nbsp;reason&nbsp;Mindcraft
&nbsp;<BR>
chose&nbsp;to&nbsp;use&nbsp;their&nbsp;own&nbsp;custom&nbsp;compiled&nbsp;package&nbsp;which&nbsp;was&nbsp;not&nbsp;compiled&nbsp;with&nbsp;&nbsp;<BR>
the&nbsp;optimizations&nbsp;that&nbsp;Red&nbsp;Hat&nbsp;compiles
&nbsp;<BR>
into&nbsp;their&nbsp;own&nbsp;packages.&nbsp;In&nbsp;particular,&nbsp;the&nbsp;Mindcraft&nbsp;package&nbsp;was&nbsp;recompiled&nbsp;&nbsp;<BR>
with&nbsp;the&nbsp;-O&nbsp;option&nbsp;rather&nbsp;than&nbsp;the&nbsp;-O2
&nbsp;<BR>
option&nbsp;that&nbsp;Red&nbsp;Hat&nbsp;uses.&nbsp;In&nbsp;other&nbsp;words,&nbsp;the&nbsp;compiled&nbsp;binary&nbsp;was&nbsp;compiled&nbsp;&nbsp;<BR>
with&nbsp;options&nbsp;that&nbsp;would&nbsp;generally&nbsp;make&nbsp;it
&nbsp;<BR>
noticably&nbsp;slower&nbsp;than&nbsp;the&nbsp;one&nbsp;shipped&nbsp;by&nbsp;Red&nbsp;Hat&nbsp;Software.&nbsp;To&nbsp;quote&nbsp;the&nbsp;GNU&nbsp;&nbsp;<BR>
&quot;gcc&quot;&nbsp;documentation&nbsp;regarding&nbsp;the
&nbsp;<BR>
&quot;-O2&quot;&nbsp;option:&nbsp;
&nbsp;<BR>
&nbsp;&nbsp;&nbsp;&nbsp;As&nbsp;compared&nbsp;to&nbsp;`-O',&nbsp;this&nbsp;option&nbsp;increases&nbsp;both&nbsp;compilation&nbsp;time&nbsp;and&nbsp;the&nbsp;&nbsp;<BR>
performance&nbsp;of&nbsp;the&nbsp;generated&nbsp;code.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
For&nbsp;a&nbsp;fair&nbsp;test,&nbsp;either&nbsp;the&nbsp;Red&nbsp;Hat&nbsp;provided&nbsp;Samba&nbsp;should&nbsp;have&nbsp;been&nbsp;used,&nbsp;or&nbsp;&nbsp;<BR>
if&nbsp;a&nbsp;'performance&nbsp;tweak'&nbsp;was&nbsp;needed&nbsp;to
&nbsp;<BR>
insure&nbsp;that&nbsp;both&nbsp;systems&nbsp;were&nbsp;optimized,&nbsp;the&nbsp;&quot;egcs&quot;&nbsp;C&nbsp;compiler&nbsp;(a&nbsp;standard&nbsp;&nbsp;<BR>
part&nbsp;of&nbsp;Red&nbsp;Hat&nbsp;Linux)&nbsp;should&nbsp;have&nbsp;been
&nbsp;<BR>
used&nbsp;with&nbsp;the&nbsp;-mpentiumpro&nbsp;option&nbsp;to&nbsp;optimize&nbsp;the&nbsp;resulting&nbsp;code&nbsp;to&nbsp;take&nbsp;&nbsp;<BR>
advantage&nbsp;of&nbsp;Pentium&nbsp;II&nbsp;features.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
2)&nbsp;Inappropropriate&nbsp;use&nbsp;of&nbsp;the&nbsp;widelinks=no&nbsp;option:&nbsp;Andrew&nbsp;Tridgell&nbsp;of&nbsp;the&nbsp;&nbsp;<BR>
Samba&nbsp;team&nbsp;reports&nbsp;that&nbsp;setting
&nbsp;<BR>
&quot;widelinks=no&quot;&nbsp;option&nbsp;will&nbsp;noticably&nbsp;slow&nbsp;performance&nbsp;of&nbsp;the&nbsp;Samba&nbsp;server.&nbsp;&nbsp;<BR>
The&nbsp;&quot;widelinks=no&quot;&nbsp;option&nbsp;is&nbsp;a&nbsp;security
&nbsp;<BR>
feature&nbsp;intended&nbsp;for&nbsp;mixed-use&nbsp;systems,&nbsp;rather&nbsp;than&nbsp;for&nbsp;dedicated&nbsp;server&nbsp;use.&nbsp;&nbsp;<BR>
For&nbsp;dedicated&nbsp;server&nbsp;use&nbsp;it&nbsp;is&nbsp;unnecessary,
&nbsp;<BR>
since&nbsp;Windows&nbsp;clients&nbsp;are&nbsp;unable&nbsp;to&nbsp;set&nbsp;filesystem&nbsp;links&nbsp;on&nbsp;a&nbsp;Samba&nbsp;server&nbsp;&nbsp;<BR>
(only&nbsp;interactive&nbsp;Linux&nbsp;users&nbsp;can&nbsp;do&nbsp;so).&nbsp;
&nbsp;<BR>

&nbsp;<BR>
Please&nbsp;note&nbsp;that&nbsp;Jeremy&nbsp;Allison&nbsp;of&nbsp;the&nbsp;Samba&nbsp;team&nbsp;reports&nbsp;no&nbsp;queries&nbsp;on&nbsp;the&nbsp;&nbsp;<BR>
Samba&nbsp;mailing&nbsp;lists&nbsp;or&nbsp;at&nbsp;the&nbsp;Samba&nbsp;home
&nbsp;<BR>
page&nbsp;that&nbsp;could&nbsp;have&nbsp;been&nbsp;about&nbsp;tuning&nbsp;Samba&nbsp;for&nbsp;use&nbsp;on&nbsp;large&nbsp;file&nbsp;servers&nbsp;&nbsp;<BR>
during&nbsp;the&nbsp;time&nbsp;period&nbsp;in&nbsp;question.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
D.&nbsp;Apache&nbsp;Configuration&nbsp;Problems&nbsp;
&nbsp;<BR>

&nbsp;<BR>
1)&nbsp;Inappropriate&nbsp;compilation:Again,&nbsp;for&nbsp;some&nbsp;unknown&nbsp;reason&nbsp;Mindcraft&nbsp;chose&nbsp;&nbsp;<BR>
to&nbsp;re-compile&nbsp;Apache&nbsp;rather&nbsp;than
&nbsp;<BR>
use&nbsp;the&nbsp;default&nbsp;Red&nbsp;Hat&nbsp;version.&nbsp;This&nbsp;makes&nbsp;it&nbsp;difficult&nbsp;to&nbsp;compare&nbsp;&quot;out&nbsp;of&nbsp;&nbsp;<BR>
box&quot;&nbsp;performance.&nbsp;If&nbsp;the&nbsp;purpose&nbsp;was&nbsp;to
&nbsp;<BR>
improve&nbsp;performance,&nbsp;using&nbsp;&quot;egcs&quot;&nbsp;to&nbsp;recompile&nbsp;it&nbsp;with&nbsp;the&nbsp;-mpentiumpro&nbsp;&nbsp;<BR>
optimizations&nbsp;would&nbsp;have&nbsp;accomplished
&nbsp;<BR>
much&nbsp;more.&nbsp;Some&nbsp;Apache&nbsp;experts&nbsp;claim&nbsp;as&nbsp;much&nbsp;as&nbsp;30%&nbsp;improvement&nbsp;by&nbsp;compiling&nbsp;&nbsp;<BR>
Apache&nbsp;with&nbsp;the&nbsp;Pentium&nbsp;Pro
&nbsp;<BR>
optimizations.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
2)&nbsp;Inappropriate&nbsp;settings&nbsp;for&nbsp;StartServers&nbsp;and&nbsp;SpareServers:&nbsp;If&nbsp;you&nbsp;are&nbsp;&nbsp;<BR>
running&nbsp;an&nbsp;enterprise&nbsp;server&nbsp;expecting&nbsp;150
&nbsp;<BR>
or&nbsp;more&nbsp;simultaneous&nbsp;connections,&nbsp;you&nbsp;would&nbsp;set&nbsp;StartServers&nbsp;to&nbsp;150&nbsp;or&nbsp;&nbsp;<BR>
greater.&nbsp;It's&nbsp;not&nbsp;as&nbsp;if&nbsp;there&nbsp;is&nbsp;a&nbsp;shortage&nbsp;of
&nbsp;<BR>
memory,&nbsp;after&nbsp;all.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
3)&nbsp;Inadequate&nbsp;information&nbsp;to&nbsp;duplicate&nbsp;the&nbsp;test:Access&nbsp;to&nbsp;copies&nbsp;of&nbsp;the&nbsp;httpd&nbsp;&nbsp;<BR>
configuration&nbsp;files&nbsp;is&nbsp;needed&nbsp;to&nbsp;further
&nbsp;<BR>
duplicate&nbsp;the&nbsp;test&nbsp;and&nbsp;diagnose&nbsp;the&nbsp;problems&nbsp;found&nbsp;there.&nbsp;For&nbsp;example,&nbsp;the&nbsp;&nbsp;<BR>
&quot;collapse&quot;&nbsp;of&nbsp;the&nbsp;Apache&nbsp;server&nbsp;described
&nbsp;<BR>
therein&nbsp;is&nbsp;counter&nbsp;to&nbsp;what&nbsp;every&nbsp;other&nbsp;performance&nbsp;survey&nbsp;of&nbsp;Apache&nbsp;has&nbsp;&nbsp;<BR>
found,&nbsp;which&nbsp;is&nbsp;that&nbsp;Apache&nbsp;reaches&nbsp;a
&nbsp;<BR>
maximum&nbsp;and&nbsp;then&nbsp;mildly&nbsp;degrades&nbsp;as&nbsp;overload&nbsp;continues.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
A&nbsp;search&nbsp;of&nbsp;the&nbsp;DejaNews&nbsp;archives&nbsp;shows&nbsp;that&nbsp;a&nbsp;&quot;<A HREF="mailto:[e-mail]"">[e-mail]"</A>&nbsp;&nbsp;<BR>
posted&nbsp;a&nbsp;query&nbsp;about&nbsp;this&nbsp;&quot;collapse&quot;&nbsp;problem.
&nbsp;<BR>
Unfortunately,&nbsp;it&nbsp;does&nbsp;not&nbsp;give&nbsp;enough&nbsp;information&nbsp;to&nbsp;make&nbsp;any&nbsp;sort&nbsp;of&nbsp;&nbsp;<BR>
diagnosis&nbsp;possible&nbsp;(probably&nbsp;why&nbsp;it&nbsp;recieved&nbsp;no
&nbsp;<BR>
meaningful&nbsp;responses).&nbsp;
&nbsp;<BR>

&nbsp;<BR>
III.&nbsp;Non-Technical&nbsp;Problems&nbsp;with&nbsp;Study
&nbsp;<BR>

&nbsp;<BR>
The&nbsp;overall&nbsp;tone&nbsp;of&nbsp;the&nbsp;report&nbsp;was&nbsp;rather&nbsp;unfortunate.&nbsp;For&nbsp;example:&nbsp;
&nbsp;<BR>

&nbsp;<BR>
We&nbsp;posted&nbsp;notices&nbsp;on&nbsp;various&nbsp;Linux&nbsp;or&nbsp;Apache&nbsp;newsgroups&nbsp;and&nbsp;received&nbsp;no&nbsp;&nbsp;<BR>
relevant&nbsp;responses&nbsp;
&nbsp;<BR>

&nbsp;<BR>
This&nbsp;implies&nbsp;that&nbsp;they&nbsp;made&nbsp;multiple&nbsp;requests&nbsp;for&nbsp;help&nbsp;(&quot;notices&quot;)&nbsp;and&nbsp;that&nbsp;&nbsp;<BR>
the&nbsp;Linux&nbsp;community,&nbsp;in&nbsp;violation&nbsp;of&nbsp;its
&nbsp;<BR>
reputation&nbsp;for&nbsp;excellence&nbsp;in&nbsp;technical&nbsp;support,&nbsp;dropped&nbsp;the&nbsp;ball.&nbsp;In&nbsp;&nbsp;<BR>
actuality,&nbsp;exhaustive&nbsp;searching&nbsp;of&nbsp;the&nbsp;DejaNews
&nbsp;<BR>
archives&nbsp;by&nbsp;a&nbsp;team&nbsp;of&nbsp;Linux&nbsp;users&nbsp;has&nbsp;identified&nbsp;one&nbsp;single&nbsp;message&nbsp;as&nbsp;a&nbsp;&nbsp;<BR>
possible&nbsp;candidate&nbsp;for&nbsp;these&nbsp;&quot;notices&quot;.&nbsp;It&nbsp;was
&nbsp;<BR>
cross-posted&nbsp;under&nbsp;a&nbsp;different&nbsp;title&nbsp;between&nbsp;a&nbsp;Linux&nbsp;newsgroup&nbsp;and&nbsp;the&nbsp;Apache&nbsp;&nbsp;<BR>
newsgroup,&nbsp;and&nbsp;recieved&nbsp;a&nbsp;reply&nbsp;which
&nbsp;<BR>
basically&nbsp;requested&nbsp;more&nbsp;information&nbsp;(albeit&nbsp;in&nbsp;a&nbsp;rather&nbsp;impatient&nbsp;manner).&nbsp;&nbsp;<BR>
If&nbsp;the&nbsp;requested&nbsp;information&nbsp;was&nbsp;ever
&nbsp;<BR>
supplied,&nbsp;it&nbsp;was&nbsp;not&nbsp;ever&nbsp;posted&nbsp;to&nbsp;the&nbsp;newsgroup&nbsp;where&nbsp;other&nbsp;members&nbsp;of&nbsp;the&nbsp;&nbsp;<BR>
Linux&nbsp;community&nbsp;could&nbsp;add&nbsp;their&nbsp;own
&nbsp;<BR>
comments.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
The&nbsp;report&nbsp;is&nbsp;full&nbsp;of&nbsp;similar&nbsp;statements&nbsp;which&nbsp;could&nbsp;be&nbsp;interpreted&nbsp;as&nbsp;&nbsp;<BR>
showing&nbsp;partiality&nbsp;or&nbsp;being&nbsp;inappropriate.&nbsp;Given
&nbsp;<BR>
the&nbsp;fact&nbsp;that&nbsp;this&nbsp;study&nbsp;was&nbsp;funded&nbsp;by&nbsp;Microsoft&nbsp;Corporation,&nbsp;Mindcraft&nbsp;&nbsp;<BR>
should&nbsp;have&nbsp;taken&nbsp;special&nbsp;care&nbsp;to&nbsp;protect&nbsp;their
&nbsp;<BR>
reputation&nbsp;by&nbsp;appearing&nbsp;as&nbsp;impartial&nbsp;as&nbsp;possible.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
IV.&nbsp;Conclusions&nbsp;
&nbsp;<BR>

&nbsp;<BR>
We&nbsp;welcome&nbsp;the&nbsp;opportunity&nbsp;to&nbsp;work&nbsp;with&nbsp;Mindcraft&nbsp;to&nbsp;resolve&nbsp;these&nbsp;glaring&nbsp;&nbsp;<BR>
issues&nbsp;and&nbsp;produce&nbsp;a&nbsp;quality&nbsp;analysis.&nbsp;The
&nbsp;<BR>
Linux&nbsp;community&nbsp;is&nbsp;extremely&nbsp;interested&nbsp;in&nbsp;a&nbsp;fair,&nbsp;impartial&nbsp;analysis&nbsp;of&nbsp;the&nbsp;&nbsp;<BR>
performance&nbsp;of&nbsp;Linux&nbsp;as&nbsp;compared&nbsp;with
&nbsp;<BR>
competitors.&nbsp;Such&nbsp;analysis&nbsp;has&nbsp;been&nbsp;very&nbsp;valuable&nbsp;in&nbsp;the&nbsp;past,&nbsp;and&nbsp;promises&nbsp;&nbsp;<BR>
to&nbsp;be&nbsp;equally&nbsp;valuable&nbsp;to&nbsp;the&nbsp;technical
&nbsp;<BR>
advancement&nbsp;of&nbsp;Linux&nbsp;in&nbsp;the&nbsp;future.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
Unfortunately,&nbsp;given&nbsp;the&nbsp;severe&nbsp;technical&nbsp;shortcomings&nbsp;of&nbsp;this&nbsp;report,&nbsp;we&nbsp;&nbsp;<BR>
must&nbsp;respectfully&nbsp;request&nbsp;that&nbsp;Mindcraft
&nbsp;<BR>
withdraw&nbsp;it&nbsp;until&nbsp;such&nbsp;issues&nbsp;are&nbsp;resolved.&nbsp;A&nbsp;report&nbsp;which&nbsp;is&nbsp;perceived&nbsp;to&nbsp;be&nbsp;&nbsp;<BR>
biased&nbsp;is&nbsp;not&nbsp;in&nbsp;the&nbsp;best&nbsp;interests&nbsp;of&nbsp;either
&nbsp;<BR>
Mindcraft&nbsp;or&nbsp;its&nbsp;client,&nbsp;and&nbsp;may&nbsp;in&nbsp;fact&nbsp;be&nbsp;embarrasing&nbsp;to&nbsp;both.&nbsp;
&nbsp;<BR>

&nbsp;<BR>
Eric&nbsp;Lee&nbsp;Green
&nbsp;<BR>
Last&nbsp;modified:&nbsp;Fri&nbsp;Apr&nbsp;16&nbsp;12:52:04&nbsp;EDT&nbsp;1999&nbsp;&nbsp;<BR>
&nbsp;<BR>
--&nbsp;<BR>
※&nbsp;来源:·BBS&nbsp;水木清华站&nbsp;bbs.net.tsinghua.edu.cn·[FROM:&nbsp;202.118.66.88]&nbsp;&nbsp;<BR>


<CENTER><H1>BBS水木清华站∶精华区</H1></CENTER>



</BODY>

</HTML>