RE: [hlds_linux] Re: [OT} Be ready--SCO to sue everything Linux???

[Stan Hoeppner]

> In any case, I guess this dashes my dream of running 500 instances of
> HLDS on the 390 I have at the office ;->


That, and the fact that the S390 processing complex wouldn't have enough
horsepower to run 500 HLDS servers (unless they were like 4 player servers).
And, assuming at least 64MB per HLDS instance, I'm guessing that your S390
wouldn't have enough physical memory either.  Last spec I read showed the
S390 maxing out at 32GB physical memory.

Oh, and btw, there is the *small* matter of getting an i386 binary running
on an S390. ;)

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Re: Be ready--SCO to sue everything Linux???

[Stan Hoeppner]

Here it comes, the ubiquitous:

I told you so.  ;)

$1 Billion?  You've got to be fucking kidding me

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX


> -Original Message-
> From: Will Day [mailto:[EMAIL PROTECTED]
> Sent: Thursday, March 06, 2003 9:33 PM
> To: [EMAIL PROTECTED]
> Subject: [hlds_linux] Re: Be ready--SCO to sue everything Linux???
>
>
> A short time ago, at a computer terminal far, far away, Stan
> Bubrouski wrote:
> >scratch and therefore is not subjet to any of their patents.
>  The article
> >mentions lawyers names and what not, but they are consultants as they
> >ahve delt with intellectual property cases before, SCO has hired them
> >to consult o nteh licensing terms it would seem.  Not for the express
> >purpose of suing anyone.
>
> fyi:
>
>SCO sues Big Blue over Unix, Linux
>http://news.com.com/2100-1016-991464.html
>
> --
> Will Day  Those who would give up essential
> Liberty, to
> [EMAIL PROTECTED]   purchase a little temporary Safety,
> deserve neither
> UNIX System ProgrammerLiberty nor Safety.
> Atlanta, GA - Benjamin Franklin, Penn.
> Assembly, Nov. 11, 1755
> ___
> To unsubscribe, edit your list preferences, or view the list
> archives, please visit:
> http://list.valvesoftware.com/mailman/listinfo/hlds_linux
>
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] CS-server behind router

[Stan Hoeppner]

> How does the won/masterserver get the port? Is the "sending-port" used
> (which is after NAT not the port the server is listening)?

Valve would have the authoritative answer on this.

If observation of the behavior of the WON servers is a correct indicator,
then yes, the WON server strips and uses the source port from the packet
that arrives at the WON server.

As to the issue of binding an IP to HLDS that doesn't exist as bound to a
local ethernet interface, I can't see how this would work.  You will most
likely get an error.

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] CS-server behind router

[Stan Hoeppner]

> my thoughts:
> IP:PORT from the network-package AFTER NAT
> for example my server has ip 192.168.1.3, port 27015. Now the
> router get's
> the package and changes the ip to my online-ip and uses a random port.
> So the server apears at the masterlist, but with a wrong
> port. So connecting
> via masterlist doesn't work.
> When the router doesn't change the port, all is fine and
> connection works
> via masterlist.
>
> Would be great if someone who KNOWS how it works can give me
> some hints :)


You've hit the nail on the head.  To fix this problem, you'll have to force
your NAT device to leave the port intact after NAT'ing the IP address.
Depending on the firmware of the router, you may or may not be able to do
this (if it's a hardware router).  If your NAT device is a PC with Linux or
another OS, you should be able to fix this problem with a rule.

Please respond with full details of your NAT device, and we'll see what we
can come up with for you.

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Be ready--SCO to sue everything Linux???

[Stan Hoeppner]

> hrmm, there's nothing in there about suing linux users, nor is there
> anything that even remotely closely resembles an urge to do so.

Ok, before the next person says this same thing, look carefully at this part
of the press release:

"The appointment of David Boies and the law firm of Boies, Schiller and
Flexner to help research and advise SCO on the company's intellectual
property."


Now, the law firm SCO hired, and the specific attorney mentioned, David
Boies, are strictly TRIAL ATTORNEYS.  That means that they TRY cases in
courtrooms.  They were hired to gather evidence of infringement of SCO's
Unix copyrights in the Linux "world", and to go to court and convince a
judge that Linux, or parts of it, or applications typcially bundled with it
(since Linux itself is really only the kernel, not the entire operating
system/environment) infringe on the SCO Unix patents in some way.

You can be absolutely sure from this that SCO will be sueing someone,
because again, that's all David Boies does.  And that someone will be the
Linux community, although specifically who we don't know yet.  It could be
Linus.  It could be Linux distributors (Red Hat, SuSe, et al).  It could be
corporations with big pockets who are using Linux in their business.  It
could be end users...

Oh, BTW, Boies defended Napster when the RIAA sued them.  Additionally, he
argued on behalf of the US Attorney General's Office against Microsoft in
its antitrust case.

The point is, one doesn't hire David Boies unless one wants to sue someone,
with the case typically ending up in a US Federal Court, or the US Supreme
Court.  One only hires David Boies if they have a big target of a lawsuit in
mind.  Boies doesn't handle small potatoes...

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

[hlds_linux] Be ready--SCO to sue everything Linux???

[Stan Hoeppner]

Just caught wind of this:

http://ir.sco.com/ReleaseDetail.cfm?ReleaseID=99965

It looks like SCO is gearing up to sue everyone using Linux into paying for
Unix licenses, since SCO owns the Unix source code.  If you look at this:
http://www.sco.com/scosource/unixtree/unixhistory01.html, it appears that
SCO and their new attorney David Boies will undoubtedly, eventually, try to
convince the US courts that Linux infringes on SCO's Unix
patents/copyrights/APIs.  (Recall the hanging chad shit in Florida in the
2000 presidential election?  Boies represented Al Gore, and argued for him
in the US Supreme Court.)

If SCO succeeds in this legal game, they'll then pull an RIAA job, and
attempt to force all Linux users to pay licensing fees

You heard it here first (unless someone mentioned this in one of the 90 some
hlds emails I deleted earlier without reading ;).

StanTheMan

rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

[hlds_linux] Aimbot influx on your server lately?

[Stan Hoeppner]

Anyone else notice an influx of aimbots on their servers recently?  Like in
the last week or so?

Our admin mailing list and IRC channel have been pretty quiet lately, until
this past weekish period.  All of a sudden, we're banning aimbotters
left/right, emails are flying among admins, and people are hopping on IRC to
scream for admin help with a h4x0r.

To top it off, I connected to one of our two servers at 2:15am CST Saturday,
as a spectator.  I noticed an AWP whoring it up, and watched for a bit.
While doing this (about 3 minutes), the server went from full at 20 players,
to 8 players.  Over half the peeps left the server

I've never seen this happen on one of our servers before.  We have admins on
all the time spectating, sometimes two or 3 at a time.  Yet, I've never seen
that kind of mass disconnect-age before.

StanTheMan
rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] CS: A few gameplay issues I'd like fixed

[Stan Hoeppner]

> And in the mean time you're eff'ed in the ear by higher
> pinging players
> and their magic curving bullets.  We now know who shot Kennedy: Mr.
> Laggers.

ROFL!  Good one Eric.

StanTheMan
rcon admin at:
Beer for Breakfast servers
   66.111.111.66:27015 (CS multi-map)
   66.111.111.66:27016 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] How *not* to water-cool your computer(s)

[Stan Hoeppner]

> http://www.avforums.com/forums/showthread.php?threadid=56924
>
> Useful information for those with heat issues in 1U cases and are
> thinking about water cooling as a solution.


I think he lied during forum registration.  He's got to be from Poland, not
Ireland, to do something like that.  Then again, Irish are known for their
*cough* sense of humor, and I'd bet a paycheck it's a wind up.

StanTheMan
TheHardwareFreak
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Is anti-terrorism killing my ping?

[Stan Hoeppner]

Thx for all the suggestions and help gentlemen.

Two additional questions:

1.  Why does Debian (Potato and Woody) not seem to include iputils in a
default net install?  I've had to install traceroute and tracepath manually
on all 3.  Strange these handy (and nesessary) little tools aren't part of
the default.  Maybe I missed something in "Install the base system"?

2.  I'd love to email the NOC for SW Bell, but I don't have an address.  I
emailed the standard "[EMAIL PROTECTED]" and got the canned response, and
no follow up.  Anyone have an address for SW Bell's NOC?  (Do they actually
have a NOC? lol)  Or an email address of someone who will actually
read/respond (other than a level 1 support jerk freshly hired away from
Burger King)?

StanTheMan
TheHardwareFreak
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] [OT] Xeon vs P3

[Stan Hoeppner]

> Xeons could be better.  They're not just an SMP capable P4.

Ahh, but that's exactly what the new Xeon is-- an SMP capable P4.  The only
differences are SMP and the optional L3 cache sizes (and hyperthreading,
although the latest--3Ghz--P4 has it also).  Also keep in mind that the
XeonDP doesn't have an L3 cache option, and that its L2 cache is the same
size as the P4.  So, the XeonDP *IS* merely a P4 with 2-way SMP support:

http://developer.intel.com/design/Xeon/prodbref/

And, regarding hyperthreading, is seems safe to assume that no one read the
links I posted before.  sigh...  At this point, Intel's Hyperthreading
(again, generically called Simultaneous Multithreading, or SMT) is nothing
more than marketing hype.  As Intel plainly states, Linux does not support
it, nor do any of MS's current server OSs.  Don't waste any brain cycles on
it.  The 10.0Ghz Xeon (yes, ten, equal to ~2 years) will be out before
hyperthreading enjoys *cough* widespread *cough* support anyway.  By that
time Hammer will be out, and then, well, all bets are off.  Early
indications are that Hammer will run Xeon into the ground, and will only
have serious competetion from Itanium3/4.  Can anyone here afford an Itanium
machine?  I can't.  And I'd bet that AMD will price at least one Hammer
variant in the "mere mortal" category.

The Xeon is an overpriced, under performing (as far as user expectations go)
processor.  It'll cost you ~double a comparable AthlonMP platform, and give
you little or no performance gain for server work.  From PriceScan:

AMD Athlon MP 1900+ 1.60GHz/Socket A/266 FSB/384K (Box) $147.00
AMD Athlon MP 2000+ 1.67/Socket A /266 FSB/384K (Box) $156.00
AMD Athlon MP 2200+ 1.8GHz/Socket A/ 266 FSB/384K (Box) $216.00
Tyan S2460 Tiger MP $159.00
Asus A7M266 DPA Multi Processor $209.00
Giga-Byte GA-7DPXDWP $240.71

Intel Xeon 2.0GHz/Socket 603/400 FST/512K (Box) $218.00
Intel Xeon 2.2GHz/Socket 603/400 FST/512K (Box) $218.00
Intel Xeon 2.4GHz/Socket 603/400 FST/512K (Box) $255.50
Intel Xeon 2.6GHz/Socket 603/400 FST/512K (Box) $367.11
Intel Xeon 2.8GHz/Socket 603/400 FST/512K (Box) $481.05
Intel SE7500CW2 $373.47
SuperMicro P4DCE+ $380.00
Tyan S2720GN $384.87


I'd suggest (since a suggestion is what the original question asked for,
iirc) acquiring a dual AthlonMP 2000+ based on a Tyan S2460/2466 mobo.
You'll get far better performance per clock tic than a Xeon system, and save
a ton of money.  Hell, for some applications, a dual AthlonMP 2000+ may out
perform a dual P4 Xeon 2.8Ghz.

To date, I've never owned an AMD system.  I've been all Intel.  However, any
new system I build will be AMD based, because of the phenomenal
price/performance advantage, ESPECIALLY regarding SMP systems.  Intel shot
itself in the foot when it decided to strip 2-way SMP suppport from the P4,
and force one to go Xeon.  This created an enormous cost hurdle for those of
us who build 2-way boxen from components.  Even if Intel had left regular
P4s with 2-way capability, I'd still go AthlonMP due to price/performance.

StanTheMan
TheHardwareFreak
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX


___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] [OT]Memory management- WAS Xeon vs P3

[Stan Hoeppner]

> (Somebody want to explain what that means - Stan?)
>
> http://support.microsoft.com/default.aspx?scid=KB;en-us;q247904

That article explains the windows terminal services problem thoroughly.  Why
are we discussing a microsoft OS memory management problem on this list?
Oh, maybe memory management needs some explanation:

Programs are written today in portable, high level languages (C, C++).
Programmers do not explicitly manage memory locations, as modern OS's handle
this for them.  This is because systems run many programs simultaneously
(multitasking), and thus one programmer cannot take into account what other
programs (and what physical memory locations they are using) may be running
on the system.  So programmers use virtual addresses, which the operating
system memory management code translates into physical memory addresses at
run time.  The problem described in the article above arises because the OS
comsumes memory in order to manage the virtual->physical mapping of memory
locations.  The consumed memory stores the translation tables, which map
virtual memory addresses to physical memory addresses.

Kind of a double edged sword, eh?  The more memory a system has (thus
allowing more processes to run, or larger processes, i.e. databases), the
more of that memory is consumed just to manage memory.  This is the price we
pay for progress... ;)

Oh, wasn't this thread about P4 Xeon performance (or lack thereof :)?
Answer in next post...

StanTheMan
TheHardwareFreak
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX




___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Binary Port

[Stan Hoeppner]

> I want to know who at valve I can talk to about either
> getting some binaries complied or possibly, starting a new
> MIPS and SPARC linux port which I would like to assume
> responsibility for. Does anyone here know who I could
> possibly contact at valve? Is it best to just email
> [EMAIL PROTECTED] or the actual maintainer of the linux
> source/release of HLDS?


Email Leon or Eric:  [EMAIL PROTECTED], [EMAIL PROTECTED]
They would be the ones to contact about an hlds binary port.

That said, we've discussed this many (a few?) times before.  A sparc port
effort was begun quite some time ago, but the developer at Sun apparently
dropped the ball, and it was never completed.

The previous response from Valve is that the effort required is not
proportional to the demand for other binary ports.  I.e. - High effort, low
demand.  The last word about binary ports is that there *will* be a X86-64
(AMD Opteron) port when Opteron hits the market.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> I would tend to disagree. If you rewrite it from scratch, why
> shouldn't
> you be able to benefit from concurrent execution?

It might benefit.  I just don't think that a significant portion of the
processing load could be done concurrently...maybe it could.  Maybe spawn a
thread to handle the processing for each client connection.  Maybe break up
the engine into X number of sub engines, each being a thread dedicated to
each player.  A new "sub-world" thread would be created for each player, and
they would then all communicate with a "whole-world" master process.  I
imagine this would require a few man-years of design and programming
though...

> Yes. :) Okay, not really, actually I think they just went the easy way
> and didn't want to go through the tougher design phase to make it
> concurrent.

I doubt this was the reason.  More likely is the fact that at the time they
were writing/designing this engine, windows 9x was the dominant platform
(still is, last figures I read).  Very few people were running NT, and they
hadn't even thought of a Linux port.  That came a few years later.  And win
9x obviously doesn't support preemptive scheduling (a requirement for
threading), nor does it support SMP.  Recall that Half-Life came out in,
what, 1998?  Windows 98 wasn't even out yet when HL hit the market.  And
almost no one was using NT for games, due to lack of DirectX support.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

[hlds_linux] [OT] some help installing debian on Smart2/P array

[Stan Hoeppner]

I would greatly appreciate anyone's help on this.  I don't have install
source (no burner =\) for Debian.  I've always done my installs with boot
disks and the mirrors.  And, I've never installed onto a hardware RAID
controller array (just IDE and plain SCSI).

So, the boot disks I have, "compact", don't have the smart2/P drivers.
Could someone please build me a RAW image of a smart2/P module disk, so I
can pop that puppy in this new quad box, and (hopefully) get Debian
installed?  Docs say the module needs to be in the floppy's boot
subdirectory.  Or, if you know where one is, just point me to the URL.
(Compact doesn't seem to support Debian, only red hat and caldera...)

I would be most greatful!

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Unix on Sparc CPUs (sun Station)

[Stan Hoeppner]

> Thanks a lot. Now, I'm wise and I finally know ;-))
>
> Really thank You.
>
> Mac


Hey, no problem.  You must be fairly new to the list, and missed that topic.
I try to keep some of the old stuff I think might be useful (because there
is no useful search function for the list archive, unless you download the
entire archive into a database or something, and then search it with your
own tools)

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers<http://bfb.bogleg.org/>
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX




>
> - Original Message -
> From: "Stan Hoeppner" <[EMAIL PROTECTED]>
> To: <[EMAIL PROTECTED]>
> Sent: Thursday, November 14, 2002 9:08 AM
> Subject: RE: [hlds_linux] Unix on Sparc CPUs (sun Station)
> |
> | From the list archive (actually *my* archive):
> |
> |
> | > Why couldnt one run Linux (such as mandrake) on a solaris
> box and then
> | > couldnt you run the standard?
> |
> | Computer CPUs (central processing units, or processors) are
> extremely high
> .
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> You sound pretty excited.. try not to get Woody *from* a
> compaq server..


ROFL.  Good one.

I can't wait to see my inbox tomorrow after everyone else (Deacon) chimes in
on this

/rolls eyes


StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> sorry, I wasnt that far through the thread. When I got to
> your posts about
> how it wont chop up the process I knew the answer to my
> question. sry :(


Hay, no need to apologize.  I just re-read a bit of my post there, and I
think I was a bit rude maybe...probably just tired/tired of the subject.  :\

Still, anyone who's not familiar with how programs run on SMP OSs/SMP
hardware, take a read of that Threading for Rookies doc.  It'll help you
understand (hopefully) that programs need to be specifically coded to really
take advantage of multiple CPUs.


StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Unix on Sparc CPUs (sun Station)

[Stan Hoeppner]

> I was wondering if anyone thought that compiling the hlds engine with
> -march=i686 flag would give better performance?  Maybe Stan
> the Man can
> answer that.

Leon could answer this best.  I recall that some of the older hlds_l
performance releases had some compiler optimizations, but exactly which
flags, I don't know.

Leon?


StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> So would openmosix be able to spread an hlds process over
> multiple cpus in
> one computer?


Didn't I already answer this question?  The answer is NO.  NO.  NO.  Mosix
can distribute multiple HLDS processes to different CPUs, depending on the
load each HLDS process is generating.  Mosix works with the ENTIRE HLDS
process.  It may move it from one node to another, but IT DOES NOT "SPLIT
UP" HLDS into sub HLDS parts.

Go read about program threads, then YOU answer YOUR question for US in your
next post.

Multithreading for Rookies:
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dndllpro/ht
ml/msdn_threads.asp

This document is geared toward threading with the win32 API, but it will
explain thread basics better than any of the Unix thread docs I could find
with Google (which would lose you in a second).

I'm pretty sure that this whole HLDS and threading nonsense has come up on
one of the two HLDS lists before (if not both), and I've posted this same
link to the "Multithreading for Rookies" M$ doc before.  Maybe that was only
the w32 hlds list.

Anyway, please, please, anyone who doesn't understand multiprocessing and
threading, read this doc.

/me tired of ranting on this subject once a month (it seems ;)

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

I'd like to know how Ryan came up with 50 milliseconds of latency for GiGE.
As m0gely points out, GiGE has a latency measured in *micro* seconds
(millionths of a second) in a switched LAN environment.

Hell, 100BaseTX and 10BaseT have latency in the lower microsecond range
(which is obviously sub millisecond).  Here are ping results over 100BaseT
(half-duplex) between 2 of my Debian boxen:

64 bytes from 192.168.100.9: icmp_seq=0 ttl=255 time=0.7 ms
64 bytes from 192.168.100.9: icmp_seq=1 ttl=255 time=0.4 ms
64 bytes from 192.168.100.9: icmp_seq=2 ttl=255 time=0.4 ms
64 bytes from 192.168.100.9: icmp_seq=3 ttl=255 time=0.4 ms
64 bytes from 192.168.100.9: icmp_seq=4 ttl=255 time=0.3 ms
64 bytes from 192.168.100.9: icmp_seq=5 ttl=255 time=0.3 ms

My half-duplex Fast Ethernet LAN pings are 700 microseconds on the high end,
and 300 microseconds on the low end (if my calculations are correct).  And
this is thru an old cheap 16 port Addtron 100TX hub.

m0gely, please clarify that 4.5 microsecond latency.  I assume he's
referring to "user_process to user_process" MPI short message latency over
Myrinet?  Don't confuse this with IP stack ping latency, which is MUCH
higher, even on Myrinet.  The reason MPI is so fast is that it actually
performs a DMA transfer between RAM segments on the nodes.  This is why MPI
short message latency is so low, and it rises as the size of the DMA
transfer increases.  Mosix/OpenMosix do NOT use MPI.  They use standard
TCP/IP.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX



> Ryan McCullough wrote:
>
> > its not bandwidth of the gigabit nic, its the latency. Lets
> say it takes
> > 50ms extra to send the data back and forth between the
> master and the
> > slaves, that adds 50ms of latency to your game. Not worth it.
>
> What if you could shave that to 4.5ms? :)
>
> "This network approach is nice because we can use a standard
> PCI slot on
> each processor node, which gives a 4.5-microsecond latency,"
> he said, as
> opposed to 90-µs latency for Gigabit Ethernet.
>
> Source:
> http://www.eet.com/at/news/OEG2002S0037
>
> --
> - m0gely
> http://quake2.telestream.com/
> Q2 | Q3A | Counter-strike
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] RE: P3 vs P4 vs Athlon was: Re: [hlds_linux] CPU Usage?

[Stan Hoeppner]

> You'll need (stan?) about 3x the airflow to keep this monster
> at a decent temperature.

I won't even guess at a ratio.  I just wouldn't run peltiers in a server at
all, especially a 1U or 2U.

> Most Peltier using overclockers tend to couple them with
> water cooling as its
> heat carrying capacity is significantly greater than air.

True.  However, the k00l'st I've seen is an evaporative cooling tower (think
"3 Mile Island"):

http://www.overclockers.com/articles389/index03.asp

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Laggy Server

[Stan Hoeppner]

Internet server, or LAN server?  If internet server, what is the size of the
internet pipe?

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX



> -Original Message-
> From: Phillip Meents [mailto:pmeents@;v4hosting.net]
> Sent: Tuesday, November 12, 2002 8:42 AM
> To: [EMAIL PROTECTED]
> Subject: [hlds_linux] Laggy Server
>
>
> Ok, im running 2 16 person CS servers, and 1 20 person CS
> server on a p4
> 1.7GHZ machine. The 2 16 person servers are empty 90% of the time. The
> people on the 20 person server experience lag spikes every couple of
> minutes. We have watched the "top" for the machine and there is no
> processor overloading and no problems with RAM, as there is 100mb of
> free ram. There is about 30GB free on the hard disk. I have also set
> maxrate to , and minrate to 2000. I have run out of
> things to check.
> The network has been fine as it is a full 100mbit connection to the
> network, and with the other 2 servers just not being used,
> its not like
> there is bandwidth problems. All of the servers are running
> on separate
> ip addresses. The machine is running on a RH 7.2 install, pretty much
> stripped to the basics. Anyone have any ideas of other stuff I could
> check. Thanks.
>
> Phillip Meents
> www.v4hosting.net
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] RE: P3 vs P4 vs Athlon was: Re: [hlds_linux] CPU Usage?

[Stan Hoeppner]

> Anyone out there using TEC modules (Thermo-electric cooling
> AKA Peltier
> device) ? I am really looking into it for a dual P3 system.

About the only people using peltiers are serious overclockers.  If you put
them inside a server (especially a small one like 1U or 2U), they'll have to
be powered from the internal power supply on the 12v rail.  Peltiers suck a
bit of power, and with a ~200-250W PSU in a 1U/2U server, your peltiers will
take a fair percentage of that available power.

Keep in mind that CPUs die very quickly if the fan on the hot side of the
peltier fails.  This could even be a potential fire hazard

I'd definitely stay away from peltiers in a server application.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: P3 vs P4 vs Athlon was: Re: [hlds_linux] CPU Usage?

[Stan Hoeppner]

> > I would never have considered running AMDs in a 1u rack
> mount regardless.
> > but if you did, I could see that heat could become a small problem,
> > especially if you cramming a few high speed HDDs into that
> small space as
> > well.  Satndard mini tower with front and back fans seems
> to run at room
> > temp in our data centre.
>
> Difference is, we cant afford to waste rack space on towers :)


Go here:

http://www.racksaver.com/products/RS1100Matrix.asp

And read this:

http://www.racksaver.com/clusters/index.asp

So, apparently RackSaver has no problem integrating 2 Athlon MPs into a 1U
chassis.  And, they offer a choice of motherboards in the 1U chassis, so
apparently they've created a few custom cooling solutions for Athlon MP in
1U, due to the different motherboard layouts.  If they're stuffing standard
19" racks full of the 1U dual Athlon MP servers to create supercomputing
clusters, I'd guarantee they've worked out any thermal issues.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> Yeah, thats wrong.  I read the FAQ before I started but I'm still not
> sure how networking works.  I guess any process I start even if its
> migrated to another system keeps the IP of its parent node?  So, data
> would come into the parent node then be forwarded to wherever the
> process is?


As far as I understand it, yes, what you state here is correct.  The I/O
(network and disk) stub of the process stays on the interactive
(home/login-in/whatever) node when the process is migrated to another node.
This is because there are two IP networks in a mosix cluster (ususally).
The first is the public interface on the interactive node.  The second is
the private interface on the interactive node which connects (usually by a
switch) to the "compute" nodes.  Typically the compute nodes do not have a
public IP, only a private cluster-only IP.  Now, you may think that IP masq
or some kind of forwarding needs to be setup on the interactive node.  I
don't believe this is correct.  Each hlds server process you will start on
the interactive node will actually bind to the public interface (which is
what you want) on the interactive node.  Then mosix will move the process to
one of the compute nodes, and handle the network traffic to/from the compute
nodes.  As far as IP networking is concerned, the hlds process is always
bound to the public interface on the interactive node.

Now maybe you're beginning to realize why mosix is so slick and c00l. ;)

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX


___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] gaming cluster.

[Stan Hoeppner]

> I'm reading all these interesting methods for gaming clusters
> and such,
> but after all the information coming in, it still seems like a single,
> multi-CPU machine would be better for these types of tasks.
> Is this not
> the case?

I would tend to agree.

> It seems like it'd be cheaper, too, judging from
> some of the
> costs I've seen here.

Agreed, until you go beyond a dual CPU machine.  That 6 CPU cluster that guy
built, if you don't attribute a $$ value to his time building it, is far
cheaper than a 4CPU x86 SMP box.  Assuming he used fast ethernet for the
switch fabric, he still spent far less money than he would have on a 4-way
box.  A current 4-way Xeon (either the older 1.6Ghz, or the new 2Ghz with
2MB L3 cache) is going to set you back a minimum of $10,000 to $20,000 US,
even buying one of the Intel white-box "to be relabeled" machines:

The chassis:
http://program.intel.com/shared/products/servers/platforms/srsh4/index.htm

Pricing on the chassis:
http://www.emicrox.com/eShop/product.aspx?cid=-1&more=1&keywords=SRSH4&pid=5
56383

Pricing on processors for this chassis:
http://www.emicrox.com/eShop/product.aspx?cid=324&pid=477846
http://www.emicrox.com/eShop/product.aspx?cid=324&pid=477850
http://www.emicrox.com/eShop/product.aspx?cid=324&pid=477853

As you can see, the chassis alone for this Intel white box 4-way, without
CPUs, memory, or HDs, is $4200 US.  The CPUs range from $1200 US to almost
$4000 US *each*.

If you look into a 4-way box from Dell, HPQ, or IBM, you're looking at a
$20,000 to $25,000 MINIMUM with all 4 CPUs.  Example:  IBM's xSeries 360
base configuration with 2 CPUs starts at $15,799:

http://www-132.ibm.com/webapp/wcs/stores/servlet/CategoryDisplay?catalogId=-
840&storeId=1&categoryId=2344267&langId=-1&dualCurrId=73

I configured an X360 with 4 1.6Ghz Xeon MP 1MB L3 cache, 2GB RAM, 3 18GB HDs
(no RAID), and the total came to $28,122 US.

I doubt the cluster guy put much more than $5,000 US into that cluster
design.  So, as far as cost, the cluster is far cheaper than a 4-way SMP
(although it will have performance hits for FPS servers during process
migration).

So, most bang for the buck will be multiples of 2-way SMP boxen, with or
without clustering.  But, most of you already knew that. ;)

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX


___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Half-Life Server Graphing

[Stan Hoeppner]

> Stan, the numbers along the bottom of the graph are times of
> the day. It
> becomes more clear when you click the graph and the entire
> MRTG history page
> shows up. It's a standard MRTG x axis quantity.
>
> Edge100x


Ahh.  Thx.  Didn't realize the image was linked, heh.

Stan
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] New Security Modules -November 4

[Stan Hoeppner]

Well apparently it affected your ability to cheat enough that it prompted
you to send this email.  Just wait till Valve improves it.

Soon, you and your cheating buddies will be out of business.  What will you
do then?  Attempt to mail flood every admin on this list, Karl?  Or, wait,
is Karl one of your gmx buddies?

"You all look the same to me."

Received: from mx0.gmx.net (SENTRY [192.168.100.4]) by
ramius.hardwarefreak.com with SMTP (Microsoft Exchange Internet Mail Service
Version 5.5.2650.21)
id Q5LHWYCY; Mon, 2 Sep 2002 02:14:54 -0500
Received: (qmail 21583 invoked by uid 0); 2 Sep 2002 00:02:52 -
Date: Mon, 2 Sep 2002 02:02:51 +0200 (MEST)
From: Karl Buck <[EMAIL PROTECTED]>

Received: from mail.gmx.net (mail.gmx.net [213.165.64.20])
by list.valvesoftware.com (Postfix) with SMTP id 6EEF72F1B
for <[EMAIL PROTECTED]>; Mon,  4 Nov 2002 17:34:53
-0800 (PST)
Received: (qmail 11127 invoked by uid 0); 5 Nov 2002 01:33:39 -
Received: from pd9ea0a53.dip.t-dialin.net (HELO rob2) (217.234.10.83)
  by mail.gmx.net (mp013-rz3) with SMTP; 5 Nov 2002 01:33:39 -
Message-ID: <001901c2846b$60658bf0$0100a8c0@rob2>
From: "Rob" <[EMAIL PROTECTED]>

Stan



> -Original Message-
> From: Rob [EMAIL PROTECTED]
> Sent: Monday, November 04, 2002 7:34 PM
> To: [EMAIL PROTECTED]
> Subject: Re: [hlds_linux] New Security Modules -November 4
>
>
> Hello,
>
> I tested it with my "testWH" ... well, ok, only for one or 2
> seconds I can
> see the enemy ... but thats enough, if it's the second before
> he comes in my
> viewing area ... Thats the second that I have to aim ant the
> enemy not ...
> :(
>
> The WH blocking must be more accurate ... this fucking
> working second must
> be eleminated ...
>
> BR Rob

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Help need For Linux Server

[Stan Hoeppner]

> Yeah, but learning screen is a very good idea.  There's
> really no reason
> to learn about ampersands and all that :)


Well, actually, I use & to start my Linux seti processes in the background,
since they are non-interactive.  And, yes, I learned about the & before
screen.  ;)

Stan
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] [OT] online stats

[Stan Hoeppner]

Personally, I kinda like this one (for CS):

http://bfb.bogleg.org/status.pl?s=bfb-cs2

;)

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX



> If so, please let me know! I'm totally shite at coding and
> would love to get
> hold of something like this!
>
> Thanks in advance for any help received.
>
> Jay.
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Help need For Linux Server

[Stan Hoeppner]

John, you should really look into screen.  Using your method, you'll not
have access to the hlds console.  Starting hlds with screen is much more
elegent, especially if you run more than one hlds server on a given host.

StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX



> You have to run it as a background process.  On the command
> line that you
> are entering ./hlds_run   , add a "&" at
> the end of the
> line.
>
> So:
> ./hlds_run -game tfc +maxplayers 20
>
> Should become:
> ./hlds_run -game tfc +maxplayers 20 &
>
> The other alternative, is to launch using something like
> halfd, which will
> restart the server when it crashes as well.
___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] Half-Life Server Graphing

[Stan Hoeppner]

> yeah works fine in combination with qstat. had a quick setup and
> I'll enhance the site with cpu, ram and nic stats due to this
> topic ;-)
>
> http://www.igazo.de/mrtg
>
> P.S. I prefer mrtg's graphs (rrdtools graphs are ugly :-)))


Ok, the URL points to a graph.  The vertical legend of the graph represents
# of players.  What is the horizontal legend (there isn't one)?  What type
of data is in the horizontal?  Why is the horizontal data split into two
sections?  Why does the horizontal start at 10 going to 24, and then start
over again at 0 going to 19?  How is one to understand that graph, when one
of the two data type represented is not given, and the on top of that, split
into two unrelated sections?

Graphs are meant to represent data, when the data itself, for whatever
reason, is not easy to digest in raw form.  Graphs, therefore, can be
misleading.  A previous graph posted here, led Prodigy to believe that HLDS
was a threaded app.

Just goes to show why graphs are so popular.  You can create one to
mis-represent any kind of data.  Maybe this is why so many graphs get
displayed during U.S. Senate debates. ;)


StanTheMan
TheHardwareFreak
http://www.hardwarefreak.com
rcon admin at:
Beer for Breakfast servers
   209.41.98.2:27016 (CS multi-map)   209.41.98.2:27015 (DoD)
   209.41.98.2:27017 (CS militia/dust2)Dallas, TX

___
To unsubscribe, edit your list preferences, or view the list archives, please visit:
http://list.valvesoftware.com/mailman/listinfo/hlds_linux

RE: [hlds_linux] NOW: HLD is NOT threaded -- WAS: Half-Life Server Graphing

[Stan Hoeppner]

> And yes, a single HLDS server IS running on two separate
> CPU's.

It is running on both CPUs, but NOT CONCURRENTLY.  What the graph is showing
(due to the data capturing method, I assume) is that the single HLDS process
is being bounced from one CPU to another.  As this bouncing can occur every
few microseconds, the data capture method thus isn't reflecting reality.

> This is
> because of the linux kernel.

Correct.  The Linux kernel, just like most SMP OS kernels, will
automatically migrate a "hungry" single thread process from a busy CPU to an
idle CPU.  And like most SMP OS kernels, the algorithm doesn't work very
well on a system with multiple CPUs and only 1 "hungry" process.  Even
though there is no benefit (and actually much performance harm) to doing
this, the kernel migrates the process anyway due to the load-leveling
algorithm, which says to run the hungry process on the least busy CPU.  The
harm comes from the fact that the CPU caches get flushed and reloaded each
time the process migrates from one CPU to another.  This "cache thrashing"
causes more access to main memory (which is very slow compared to L2 cache,
on the order of 1000 times slower with 1Ghz+ CPUs w/on-die L2 cache), and
eliminates the performance gained from a fast L2 cache.  If however, you
have more than one "hungry" process, each of them will tend to continue
executing on the current CPU, instead of being migrated.  This is one
downside to programming an SMP kernel that works well when heavily loaded
with multiple processes.  It falls down when running only one "hungry"
process on a computer with multiple CPUs.

> I'm wondering if  w2k/xp kernel
> will also
> allow for this?

Yes, it does.  But again, this is NOT a positive thing.  And you can see it
in realtime.  Here's a test:  Download the w32 setiathome wnnt-cmd-line
client onto an SMP NT (W2K/XP) machine:

ftp://ftp.cdrom.com/pub/setiathome/setiathome-3.03.i386-winnt-cmdline.exe

It has no install routine or graphics output, so save it somewhere.  You
launch it from the command line (hence the name) from the directory it
resides in.  This app will consume 100% of a CPU, as it is a single
execution thread (just like HLDS), and floating point heavy.  Launch it with
the -verbose switch so it shows processing status in the running window:

setiathome-3.03.i386-winnt-cmdline.exe -verbose

The first time you run it, it will ask you to create an account with your
email address, so you get credit for processing the work units.  Example:

http://setiathome.ssl.berkeley.edu/fcgi-bin/fcgi?email=stan@;hardwarefreak.de
vastation.cc&cmd=user_stats_new

Once it's running (and assuming you have no other apps eating much CPU),
fire up the task manager, and watch the CPU usage graphs.  You'll see that
it *appears* that the app is using both CPUs, but not all of either CPU.
What is actually happening is that the process is being bounced from one CPU
to the other, but the time interval in which the NT/W2K/XP kernel captures
CPU load is greater than the kernel slice in which the process executes (on
the order of 100 microseconds per kernel timeslice).  Thus, it appears that
both CPUs are being used at the same time, BECAUSE THE DATA CAPTURE METHOD
DOESN'T WORK ON A SMALL TIMESCALE (I.E. millionths of a second, which IS the
timescale the OS kernel schedules tasks to execute within).  To prove my
point, kill the process (CTRL+C).  Launch it again with the -CPU N switch:

seti303.exe -CPU 1  (I've renamed my clients because of the filename
length ;)

Now what you'll see in the task manager is that the process is being forced
(by a kernel paramater) to execute on only CPU 1 (the right pane CPU usage
graph on a 2 CPU system-- the left pane being CPU 0).  Now, keep in mind
that the MS NT kernel can be passed a runtime parameter from an application,
telling it to run that process on ONLY a specified CPU.  I am not aware of a
Linux kernel that has this functionality.  Some other flavors of Unix do
(SGI Irix comes to mind).  I'm not sure why Linus decided not to exclude
this functionality from the Linux kernel.  It could be very beneficial.
Then again, maybe it is there and I'm just unaware of it, or haven't been
able to find it (believe me, I've searched).

Now, if you reply and tell me my test method is wrong, and contiue to
support the argument that the graphs are correct, and that the app just
doesn't need all of either CPU, run this same test on a system with faster
CPUs.  You'll see the same behavior no matter how fast or slow the CPUs are,
from a Pentium 100 to a P4 3.06Ghz.  The process is a single thread
(execution stream), and can only run on one CPU at a time, JUST LIKE HLDS.
Ask Eric Smith, or Leon Hartwig.  They'll tell you the same thing.  HLDS IS
A SINGLE THREAD PROCESS.  It is serial in nature, and has no re-entrant
code, thus it cannot execute on more than one CPU at a time.

Then again, since you posted this topic already, you don't have enough
knowledge of OS kernels or prog