Very Computer

See:

        http://www.microsoft.com/backstage/column_T2_1.htm

and:

        http://www.microsoft.com/misc/backstage/solutions.htm

This is one of the best real-world case studies on why not to use NT
for servers that I have seen; sometimes you have to read between the
lines and have a good clue to see why they are making a good case
against NT, sometimes it is darn obvious.

Right, quite interesting reading.

But could someone explain objectively (on an advocacy newsgroup, yeah,
right...) to us mere mortals what the numbers mean?

Thanks.

|> Right, quite interesting reading.
|>
|> But could someone explain objectively (on an advocacy newsgroup, yeah,
|> right...) to us mere mortals what the numbers mean?

Explain, no, but a couple of observations...

|> >
|> > See:
|> >
|> >         http://www.microsoft.com/backstage/column_T2_1.htm

Should be lower case "t", http://www.microsoft.com/backstage/column_t2_1.htm
(linked to from page below)

Talks about a kind of IP "front-end" to shield users from crashing servers:

        "Until Single IP, we were just like everyone else: we never had
         a 100 percent day. Never."

Well, that says a lot, now doesn't it?  :-)

        "Just running CHKDSK on a 36GB server, which is standard after
         a crash, can take up to three hours."

|> > and:
|> >
|> >         http://www.microsoft.com/misc/backstage/solutions.htm

...Hardware
  *  Runs on Compaq Proliant 5000s and 5500s, with 4 Pentium Pro processors
     and 512 megabytes (MB) of RAM each.
...
The physical architecture behind www.microsoft.com seems surprisingly
modest. Twenty-nine servers host general Web content; 25 servers host
SQL, 6 respond to site searches; 3 service download requests along with
another 30 in distributed data centers; and 3 host FTP content. Additional
servers overseas handle some of the international load.   ...

That's over *NINETY-SIX* servers!  Maybe not all are 4-CPU/512MEG, but
still.  And to them, that's "surprisingly modest"...

|> > This is one of the best real-world case studies on why not to use NT
|> > for servers that I have seen; sometimes you have to read between the
|> > lines and have a good clue to see why they are making a good case
|> > against NT, sometimes it is darn obvious.

Big job = HUGE hardware requirements and other work-arounds for software
inadequacies?

--
Bill Stapleton                          University of Wisconsin - Milwaukee

    Web Janitor, http://www.uwm.edu/             Technical Solutions

No, it should be upper case T as linked to from
http://www.microsoft.com/backstage/default.htm.  Ooops, there we go again:
servers that aren't case sensitive and where the content creators don't
ensure consistent case result in horrible cache busting.  At least
the don't (that I noticed) do the same with an images directory name.

Quote:>Talks about a kind of IP "front-end" to shield users from crashing servers:
>    "Until Single IP, we were just like everyone else: we never had
>     a 100 percent day. Never."
>Well, that says a lot, now doesn't it?  :-)

To some degree.  More telling is their graphs showing that 95% or below
availability was common.  It appears that they were all tied up for
a long time over their inability to use common, scalable third party
solutions for this that have been around for year.  NIH runs rampent.

Heck, they didn't even need any third-party hardware to do a better
solution than what they have had for years.  Took them long enough
to find their great new technology.

I have my long quarterly Microsoft rant on this article that I have
written, but don't feel like posting it here since it is a rant, not an
analysis.

Thanks, couldn't see the forest for the trees,... well, maybe, it's just
lack of coffee ;)

So to summarize in layman talk:

To attain 100% availability *days*, (not weeks or months...) you need
lots and lots of hardware and software redundancy so that *when* (not
if) a server crashes your customers don't notice...

If your NT server goes down, chkdsk goes chugging along at  36GB / 180
mins   = 0.2GB/min (ouch!) before you can restart.

To service 228 + 12 million hits, you need 96 computers which works out
to 1 server per 2.5 million hits. Hmm...Sounds like they just replicated
the box under the developer's table. Very Modest.

Good. Very reliable, efficient and cost effective, not.

No wonder Mark Slemko was ranting.

Regards,

Wei Shun

<snip>

Quote:>         "Just running CHKDSK on a 36GB server, which is standard after
>          a crash, can take up to three hours."

> That's over *NINETY-SIX* servers!  Maybe not all are 4-CPU/512MEG, but
> still.  And to them, that's "surprisingly modest"...

</snip>

: Thanks, couldn't see the forest for the trees,... well, maybe, it's just
: lack of coffee ;)

: So to summarize in layman talk:

: To attain 100% availability *days*, (not weeks or months...) you need
: lots and lots of hardware and software redundancy so that *when* (not
: if) a server crashes your customers don't notice...

: If your NT server goes down, chkdsk goes chugging along at  36GB / 180
: mins   = 0.2GB/min (ouch!) before you can restart.

It's 4M per second - perhaps chkdsk does something really painful like
checking CRCs on the sectors themselves, so is bounded by data
transfer rate. Just being charitable, though ...

Remember, they're handling a quarter-billion hits a day, or a couple
of hits a millisecond, _and these are .ASP pages requiring a couple of
database lookups each_ - you're not going to manage that from a P90,
I'd have thought, and you might well want one server to handle each of
your T3 lines (though they use OCI2 lines of 600Mbps capacity).

Tom

: See:

:       http://www.microsoft.com/backstage/column_T2_1.htm

        Microsoft invents Single IP solution? ;-)
http://www.f5.com/index.phtml?page_id=03010000
http://www.cisco.com/warp/public/751/lodir/

--
-----------------------------------------------------------------------

Blinky lights are the essence of  |  we'll just print more."  
modern technology!                |  Caffeine underflow (brain dumped)

I loved the "dog food" bit particularly.

 - David

--
-----------------------------------------
David Joffe  djoffe at icon dot co dot za
http://www.geocities.com/SoHo/Lofts/2018/
-----------------------------------------

On Tue, 06 Oct 1998 18:22:44 +0800, Lee Wei Shun

Well I've just drank a cup, so I should be ok. ;-)

Quote:>So to summarize in layman talk:

>To attain 100% availability *days*, (not weeks or months...) you need
>lots and lots of hardware and software redundancy so that *when* (not
>if) a server crashes your customers don't notice...

I think they were shooting themselves in the foot having 96 computers,
with no automagic hot-swap going on (Did someone mention Wolfpack? ;-)
It is a lot harder to keep 96 computers up than one.

Quote:>If your NT server goes down, chkdsk goes chugging along at  36GB / 180
>mins   = 0.2GB/min (ouch!) before you can restart.

That sum surely isn't correct - they won't do a full surface scan
after a crash, surely only checking the system areas. Surely? Makes
the transfer rate even worse though...

Quote:>To service 228 + 12 million hits, you need 96 computers which works out
>to 1 server per 2.5 million hits. Hmm...Sounds like they just replicated
>the box under the developer's table. Very Modest.

The peaks will be very high, and the troughs very low. Seems excessive
though. Wonder what Netscape use at www.netscape.com. I would imagine
it would have a fair few simultaneous hits too. Wonder what will be
the hit rate at www.idsoftware.com when Quake III comes out. ;-)

Quote:>Good. Very reliable, efficient and cost effective, not.

Nice big numbers though.

That's 49 Gig of real memory.

Quote:></snip>

Regards

Anthony
===============================================================
|'All kids love log!'                                         |
|                                              Ren & Stimpy   |
===============================================================

[Profile]
Build="32.451"
FullName=""
EMailAddress=""
ReplyTo=""
Organization=""
DoAuthorization=0
SavePassword=0
UserName=""
Password=""
IsRegistered=0
IsLicensed=0
Key=""
EnableSupportMenu=0

[Servers]
NewsServer=""
MailServer=""
POPServer=""
NNTPPort=119
SMTPPort=25
POPPort=110
SMTPServerPort=25

[Groups]
LastUpdate=""
RefreshMode=0
RecordGaps=0
DemonDolt=1
MinGapCount=5
SampleMode=0
SampleCount=50
SampleDays=7
SampleUpdateRanges=1
UnkeepUnavailable=1
Sorted=1
ViewEmptyAction=32767
SampleNewCount=32767
LastFolderSelection=""

[Online]
AutoDisconnect=1
NoActivityTimeout=300
MailAutoDisconnect=1
MailNoActivityTimeout=5
SendKeepAliveMsgs=0
KeepAlivePeriod=60
XHdrOnly=0
EnablePriorityTask=1
PriorityAttachments=0
AutoView=0
RetryGetHostName=0
RetryConnectAttempts=0
RetryConnectPause=30
RetryConnectCloseWinsock=0
UseGetHostName=0
EnableMsgIdRetrieval=1
ErrorTimeout=300
LoadWinsockOnStartup32=1
HangUpOnStop=0
RasOpen=0
RasOpenPrompt=0
RasConnection=""
RasUseAny=1
RasClose=1
RasClosePrompt=1
RasUserName=""
RasPassword=""
RasConnectedTitle="Connected to "
RasRunMonitor=1
RasMonitor="rasmon.exe"

[View]
AutoOpenMode=1
ShowToolBar=1
ToolBarSize=100
ShowStatusBar=1
ProgressStyle=0
ProgressDelay=1000
InitialView=1
GroupPropTab=0
DefGroupPropTab=0
ProfileTab=0
PrefsTab=0
PostingTab=0
InboundTab=0
DisplayTab=0
MenusTab=0
URLTabItem="http"
MainWindowPosition="0 1 -1 -1 -4 -4 73 96 951 672"
BrowserPosition="2 3 -1 -1 -4 -23 0 0 716 327"
Title=""
ReturnToMessage=1
SelectedMessage="0,"
LastView="0,0,0"
Win95=0
StatusBarFont="MS Sans Serif, 80, 0, 0, 0,0,0"
HeaderInfoFont="Arial, 80, 0, 0, 0,0,0"
OldStyleBodyHeader=0

[Browser]
NormalFont="MS Sans Serif, 80, 0, 0, 0,0,0"
BoldFont="MS Sans Serif, 80, 1, 0, 0,0,0"
ScrollMargin=2
FollowUpIndent=4
MaxNestLevel=6
CollapseThreads=1
ExpandNewThread=0
ExpandFromBody=0
SubjectThreading=1
NewSubjectStartsNewThread=1
NextStaysInPane=1
MarkSkipsToNextUnread=1
WatchSkipsToNextUnread=1
SkipGroupMarksRead=1
RightButtonSelects=1
ShowUnreadOnly=0
ShowBodiesOnly=0
SortDateDescending=0
SortSizeDescending=0
ScrollLock=0
MarkReadMode=1
MarkReadReply=0
SingleKeyViewsBody=1
EnableWheelButton=1
WheelButtonPagesDown=0
ViewLaunchesOnRetrieve=0
LaunchInlineAttachments=0
BackupLimit=100
DefaultGroupView=""
DefaultFolderView=""

[Article]
UseFixedFont=0
WordWrap=0
ShowHeader=0
ShowRaw=0
ShowDividers=1
UseFixedFontRaw=0
WordWrapRaw=0
ShowHeaderRaw=1
QuotedTextMarkers=">:|"
TabSpacing=8
SaveArticleFormat=0
SaveArticleSeparator="========\n"
SaveArticleFields=2
SaveArticleFieldList="To,Newsgroups,Subject,From,Date,Cc,Bcc"
SaveArticleAppend=0
SaveArticleRaw=1
ImportUseXFolder=1
ImportCreateFolders=1
ImportFileMode=0
ImportFileFolder=""
PageDownMode=0
ShowFullAddress=0
NewsBaseFields=""
MailBaseFields="!To,!Subject,!From,!Date,Cc,Bcc"
NewsAllFields=""
MailAllFields=""
AttachedBaseFields="Newsgroups,To,Subject,From,Date,Cc,Bcc"
AttachedAllFields=""
ScrollLock=1
FilterAppleDouble=1
FileMeansAttachment=0
SaveNamedInlines=1
ContentMeansMime=1
DblClkOpensIndex=0
SkipAlternatives=1
ReplyPrefixes="re"

[Attachment]
PartitionMode=0
PartitionLines=10000
TextMode=0
FileMode="644"
NameConversionMode=0
MaxTempFiles=10
Executables="exe,bat,com,cmd"
DeleteAfterJoin=1
RequireFilenameWithTag=1
LastMimeApp=""
SaveOverwriteMode=2

[Purging]
PurgeOnNewHeaders=1
PurgeOnExit=0
ConfirmPurgeOnExit=1
PurgeByNumber=1
AlwaysPurgeUnavailable=1
CompressOnPurge=0
CompressOnExit=0
CompressPercent=20

[Message]
UseFixedFont=0
WordWrap=0
IncludeOriginalText=1
QuotePrefix=">"
FollowUpString="On %date%, %from%

it seems that my network card is not recognized

it is in a pci slot irq 9,i/o 6100-611f
in a pci slot
its an etp-4204-ep   32 bit ethernet adapter card
by unicom electric inc

any help is appreciated and thanks in advanced

        You need to know what chipset it is using.

        Docs from the vendor would be preferable.

        This is legacy hardware I presume?

--
Unix had  startmenus and tasbars before Microsoft          |||
even had a decent memory manager for DOS.                 / | \

        In search of sane PPP docs? Try http://penguin.lvcm.com