Re: Speed isn't everything, luckily for OpenBSD.

2005-07-23 Thread J.C. Roberts
On Sat, 23 Jul 2005 01:08:04 -0400, Brad [EMAIL PROTECTED] wrote:

On Fri, Jul 22, 2005 at 09:43:29PM -0700, J.C. Roberts wrote:
 On Fri, 22 Jul 2005 21:10:53 -0400, Nick Holland
 [EMAIL PROTECTED] wrote:
 
 There is just *no* way to explain just how wacked Linux looks to someone
 who is having to go from OpenBSD to Linux for some stuff at work.  Wow.
   You'd swear it was written by an unorganized mob with no central
 control or plan at all.  Oh, wait...
 
 Nick.
 
 ROTFLAMO! -If you think the Server/Desktop linux distros are bad, you
 should see some of the completely wacked linux incarnations that
 ship with custom reference boards from various chip manufacturers.

Ya but the intention is for embedded use. The end-user typically has no
interaction with such distributions. I'm not saying that's a good excuse
to make the developers lives harder though.

I *think* we agree but I'm still a bit unsure of your point?

Sure, the ASIC vendors *expect* us to know the exact incantation to
mumble over our voodoo chicken sacrifice in order to get their stuff
working but I think this mess is not very different than the whole
optimization nonsense seen so often here on the OpenBSD lists.

Whether it's a new atmel ARM SoC chip or some custom built video or
crypto ASIC, the people in the best position to provide sane defaults
are the folks that built the darn thing. In the world of linux where
there are no sane defaults and everyone is *expected* to turn a few
zillion hidden knobs, a whole lot of time/money gets wasted, again and
again and ...

JCR

--
A: Because idiots do not know how to configure their email programs.
Q: How does top-posting happen?
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing in e-mail?



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-23 Thread MikeM
On 7/22/2005 at 9:10 PM Nick Holland wrote:

| There is just *no* way to explain just how wacked Linux looks to 
| someone who is having to go from OpenBSD to Linux for some stuff 
| at work.  Wow.
| You'd swear it was written by an unorganized mob with no central
| control or plan at all.  Oh, wait...
 =

Software tends to take on the architecture of the organization that
created it.



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-23 Thread Brian
--- MikeM [EMAIL PROTECTED] wrote:

 On 7/22/2005 at 9:10 PM Nick Holland wrote:
 
 | There is just *no* way to explain just how wacked Linux looks to 
 | someone who is having to go from OpenBSD to Linux for some stuff 
 | at work.  Wow.
 | You'd swear it was written by an unorganized mob with no central
 | control or plan at all.  Oh, wait...
  =
 
 Software tends to take on the architecture of the organization that
 created it.
 

Fortunately, the group here stands fast and creates good stuff.  You have to
respect a group that will tell you straight out that you are making mistakes. 
I actually solved my little assembly problem thanks to the approach the
developers take here.

Brian
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Henning Brauer
* Kevin [EMAIL PROTECTED] [2005-07-22 09:53]:
 On 7/21/05, Lars Hansson [EMAIL PROTECTED] wrote:
  On Thu, 21 Jul 2005 12:35:27 -0500
  Kevin [EMAIL PROTECTED] wrote:
   To be blunt, because when an enterprise just needs pure unfiltered
   inter-VLAN routing, Cisco has CEF products which can route between
   interfaces at bps and pps rates unapproachable using a general purpose
   Unix OS and COTS hardware.
  You know that CEF is just a poor exscuse for the pathetic performance
  of the CPU's Cisco put in, right?
 Fine, so you don't like Cisco.

most of us just don't lik their hogwash. they sell crap.


 Substitute Raptor or Juniper or some other 
 product that can do basic Inter-VLAN routing at 100,000 packets/second
 in even their low end products,  That doesn't change the facts, just the
 brand name, the hue of the case, and maybe the reseller's profit margin.

opposed to cisco juniper ships software that at least got some kind of 
QA, and doesn't sell pathetically slow CPUs for 10s and 100s of K$, and 
doesn't impose arbitary memory limits to force you to buy new boxens 
after a way too short period of time, and and and and...

so, yes, it changes the facts dramatically.

 Regardless of whether you use the walks-on-water SysKonnect cards or
 crappy $470 Intel quad-EM cards, OpenBSD on i386 barely approaches
 half that rate, when doing nothing more than routing packets from one
 interface to another (but I'd be happy to be proven wrong on this).

100Kpps should be reachable with the right hardware right now.

there is room for optimization in OpenBSD to reach way higher 
forwarding rates.

-- 
BS Web Services, http://www.bsws.de/
OpenBSD-based Webhosting, Mail Services, Managed Servers, ...
Unix is very simple, but it takes a genius to understand the simplicity.
(Dennis Ritchie)



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Lars Hansson
On Fri, 22 Jul 2005 02:31:23 -0500
Kevin [EMAIL PROTECTED] wrote:

 Fine, so you don't like Cisco.
No sane person *likes* Cisco gear.

  Substitute Raptor or Juniper or some other 
 product that can do basic Inter-VLAN routing at 100,000 packets/second
 in even their low end products,

Dude, there are no low-end systems that reach anything even remotely close
to 100kpps, especially not Cisco ones. That's why they're called *low-end*.

  That doesn't change the facts, just the
 brand name, the hue of the case, and maybe the reseller's profit margin.

The fact is that most other vendors make better gear for less price (with
the exception of Huawei...ugh). That changes the playing field quite 
drastically.

 Regardless of whether you use the walks-on-water SysKonnect cards or
 crappy $470 Intel quad-EM cards, OpenBSD on i386 barely approaches
 half that rate, when doing nothing more than routing packets from one
 interface to another (but I'd be happy to be proven wrong on this).

(see Henning's reply)

 but I wouldn't
 try to sell my boss on replacing the big 6500 core routers with a couple of
 OpenBSD machines stuffed full of PCI-X cards.

That's exactly what some companies make and sell only it's FreeBSD and not 
OpenBSD.

 IOW, if the question is How do I transfer packets as quickly as possible
 from one broadcast domain to another, without any thought for security?,
 maybe the answer shouldn't be OpenBSD.

...then again maybe it is.

---
Lars Hansson



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Joe .
On 7/22/05, Henning Brauer [EMAIL PROTECTED] wrote:

  Regardless of whether you use the walks-on-water SysKonnect cards or
  crappy $470 Intel quad-EM cards, OpenBSD on i386 barely approaches
  half that rate, when doing nothing more than routing packets from one
  interface to another (but I'd be happy to be proven wrong on this).
 
 100Kpps should be reachable with the right hardware right now.
 
 there is room for optimization in OpenBSD to reach way higher
 forwarding rates.


Part of the problem here is that people need to know the optimizations
and unfortunately it seems like it takes either lots of research
and/or a bit of luck to get the best combo. Instead it would be great
if there was either a small addition to the faq or a separate faq that
gave simple and explicit directions in how to turn an openbsd server
into a lean mean routing machine. (Henning: Life with an OpenBSD
router?)

It would definitely help to lower the barrier of entry and then
instead of rehashing this topic you could just point people to that
doc (and perhaps throw some real world results in it for good
measure). This would help people make the case for openbsd at work to
their managers as well.

Joe



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Adam
On Fri, 22 Jul 2005 02:31:23 -0500 Kevin [EMAIL PROTECTED] wrote:

 Fine, so you don't like Cisco.  Substitute Raptor or Juniper or some
 other product that can do basic Inter-VLAN routing at 100,000 packets/
 second in even their low end products,  That doesn't change the
 facts, just the brand name, the hue of the case, and maybe the
 reseller's profit margin.
 
 Regardless of whether you use the walks-on-water SysKonnect cards or
 crappy $470 Intel quad-EM cards, OpenBSD on i386 barely approaches
 half that rate, when doing nothing more than routing packets from one
 interface to another (but I'd be happy to be proven wrong on this).

PCs aren't as slow as you think they are.  After taking the time to
tune their network stack for performance, FreeBSD 5.3 can route 1Mpps
on a high end PC.

I have no idea how many pps openbsd can handle, but obviously you don't
have to resort to cisco just to get 100,000 pps.

Adam



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Nick Holland
On Fri, Jul 22, 2005 at 08:48:11AM -0400, Joe . wrote:
 On 7/22/05, Henning Brauer [EMAIL PROTECTED] wrote:
  
  100Kpps should be reachable with the right hardware right now.
  
  there is room for optimization in OpenBSD to reach way higher
  forwarding rates.
 
 
 Part of the problem here is that people need to know the optimizations
 and unfortunately it seems like it takes either lots of research
 and/or a bit of luck to get the best combo. Instead it would be great
 if there was either a small addition to the faq or a separate faq that
 gave simple and explicit directions in how to turn an openbsd server
 into a lean mean routing machine. (Henning: Life with an OpenBSD
 router?)
 
uh, no.
I think you missed the point of Henning's comment.
This is something for developers to work on, not magic knobs for you to
twist.  If there was something worthy of putting in the FAQ about this, 
it would be in the CODE.  This isn't some other OS that shall remain 
nameless, where you are expecting to get a barely functional system and
have to twist a lot of knobs to make it run properly, OpenBSD is
supposed to have a good set of settings out of the box on the
best-tested version of the kernel.

Somehow, people seem to think there is some magic tweek we hide from 
people that will make their systems run better.  If that were the 
case, it would be set ALREADY.  Sheesh...why would we hide such a thing?
why wouldn't it be set that way if it was a always do this kind of 
thing?  The people who persue these kinds of tweeks usually are trying
to optimize their cable modem or a T1 line that wasn't coming close to
pushing the limits, anyway.  Might explain all the stupid looking fins
and funny exhaust systems I see bolted on cars...which is fine, just
don't expect the auto maker to help you...nor exect it to get you
through traffic any faster...inor get through the track any faster when
you are pushing down on the rear wheels of a front wheel drive car
*sigh*.  But I digress...
 
 It would definitely help to lower the barrier of entry and then
 instead of rehashing this topic you could just point people to that
 doc (and perhaps throw some real world results in it for good
 measure). This would help people make the case for openbsd at work to
 their managers as well.

No, lowering hte barrier involves making things Just Work without 
tweeking silly knobs.

Nick.



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Joe .
On 7/22/05, Spruell, Darren-Perot [EMAIL PROTECTED] wrote:
 
 I think, quite the opposite, that it's fine the way it is. It's not
 openbsd's fault that people fall prey to the stupid knob-tuning game and
 quite dumbly follow that line of thought. I think instead that the other
 OSes should be responsible for slapping a disclaimer on their {box, web
 page} saying something like This operating system, contrary to rational
 thinking, is not optimized for the most reasonable performance under the
 most common use cases. Instead of being functional out-of-the box, you are
 expected to re-compile critical portions of the system in order to get them
 to work to your specifications. If you don't find this behavior intuitive,
 feel free to use a more rational, completely functional operating system
 instead.
 

I agree with you completely and in a sane and rational world it would
happen just like that. Unfortunately I highly doubt we'll see any such
disclaimers though. I bet there are lots of people eager to defect and
it would definitely make things easier if such concepts were made
clear (maybe there could be a Bad OS Refugee page in the faq).



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Timothy Donahue
On Friday 22 July 2005 01:23 pm, Spruell, Darren-Perot wrote:
 From: Joe . [mailto:[EMAIL PROTECTED]

  are used to dealing with complex or unoptimized piles of crap. Part of
  encouraging people to switch should at the very least be communicating
  that there are no hidden options or that straightaway things are going
  to work as best as possible.

 I think, quite the opposite, that it's fine the way it is. It's not
 openbsd's fault that people fall prey to the stupid knob-tuning game and
 quite dumbly follow that line of thought. I think instead that the other
 OSes should be responsible for slapping a disclaimer on their {box, web
 page} saying something like This operating system, contrary to rational
 thinking, is not optimized for the most reasonable performance under the
 most common use cases. Instead of being functional out-of-the box, you are
 expected to re-compile critical portions of the system in order to get them
 to work to your specifications. If you don't find this behavior intuitive,
 feel free to use a more rational, completely functional operating system
 instead.

 DS

I'm going to have to agree with Henning, an operating system should be 
configured by the people who develop it to have sane defaults for all but the 
most unique cases and in my experience OpenBSD does an excellent job at doing 
this.  The people that have the strange corner cases where performance 
tuning will make a large difference generally also have the staff, or the 
money to pay for someone with the experience, to do the tuning.  In my mind, 
these are generally extreme cases where where throwing more processing power 
at the problem is not an efficient solution, for example huge Trading markets 
where transaction processing time counts (NYSE, NASDAQ).  

Tim Donahue



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Spruell, Darren-Perot
From: Joe . [mailto:[EMAIL PROTECTED]
  I think, quite the opposite, that it's fine the way it is. It's not
  openbsd's fault that people fall prey to the stupid 
 knob-tuning game and
  quite dumbly follow that line of thought. I think instead 
 that the other
  OSes should be responsible for slapping a disclaimer on 
 their {box, web
  page} saying something like This operating system, 
 contrary to rational
  thinking, is not optimized for the most reasonable 
 performance under the
  most common use cases. Instead of being functional 
 out-of-the box, you are
  expected to re-compile critical portions of the system in 
 order to get them
  to work to your specifications. If you don't find this 
 behavior intuitive,
  feel free to use a more rational, completely functional 
 operating system
  instead.
  
 
 I agree with you completely and in a sane and rational world it would
 happen just like that. Unfortunately I highly doubt we'll see any such
 disclaimers though. I bet there are lots of people eager to defect and
 it would definitely make things easier if such concepts were made
 clear (maybe there could be a Bad OS Refugee page in the faq).

http://www.openbsd.org/faq/faq9.html#Introduction

Read between the lines and that's what we have... ;)

DS



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Nick Holland
Spruell, Darren-Perot wrote:
 From: Joe . [mailto:[EMAIL PROTECTED]
...
 I agree with you completely and in a sane and rational world it would
 happen just like that. Unfortunately I highly doubt we'll see any such
 disclaimers though. I bet there are lots of people eager to defect and
 it would definitely make things easier if such concepts were made
 clear (maybe there could be a Bad OS Refugee page in the faq).
 
 http://www.openbsd.org/faq/faq9.html#Introduction
 
 Read between the lines and that's what we have... ;)

Heh.  Glad to see someone caught some of the between the lines stuff
in there. :).  That was a hard article to rework, so hard to not just
say, Get over yourself, your use of Linux makes you much an expert on
Unix as using Applesoft BASIC when you were ten makes you an expert on
programming

There is just *no* way to explain just how wacked Linux looks to someone
who is having to go from OpenBSD to Linux for some stuff at work.  Wow.
  You'd swear it was written by an unorganized mob with no central
control or plan at all.  Oh, wait...

Nick.



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread J.C. Roberts
On Fri, 22 Jul 2005 21:10:53 -0400, Nick Holland
[EMAIL PROTECTED] wrote:

There is just *no* way to explain just how wacked Linux looks to someone
who is having to go from OpenBSD to Linux for some stuff at work.  Wow.
  You'd swear it was written by an unorganized mob with no central
control or plan at all.  Oh, wait...

Nick.

ROTFLAMO! -If you think the Server/Desktop linux distros are bad, you
should see some of the completely wacked linux incarnations that
ship with custom reference boards from various chip manufacturers.

JCR

--
A: Because idiots do not know how to configure their email programs.
Q: How does top-posting happen?
A: Because it messes up the order in which people normally read text.
Q: Why is top-posting such a bad thing?
A: Top-posting.
Q: What is the most annoying thing in e-mail?



Re: Speed isn't everything, luckily for OpenBSD.

2005-07-22 Thread Brad
On Fri, Jul 22, 2005 at 09:43:29PM -0700, J.C. Roberts wrote:
 On Fri, 22 Jul 2005 21:10:53 -0400, Nick Holland
 [EMAIL PROTECTED] wrote:
 
 There is just *no* way to explain just how wacked Linux looks to someone
 who is having to go from OpenBSD to Linux for some stuff at work.  Wow.
   You'd swear it was written by an unorganized mob with no central
 control or plan at all.  Oh, wait...
 
 Nick.
 
 ROTFLAMO! -If you think the Server/Desktop linux distros are bad, you
 should see some of the completely wacked linux incarnations that
 ship with custom reference boards from various chip manufacturers.

Ya but the intention is for embedded use. The end-user typically has no
interaction with such distributions. I'm not saying that's a good excuse
to make the developers lives harder though.