Unix Programming - A discussion of Linux and BSD - about scaliability, consistency, robustness and IP-St

Author

A discussion of Linux and BSD - about scaliability, consistency, robustness and IP-St

Penang

2004-10-29, 3:58 am

The other day, on slashdot, someone mentioned that BSD's IP/Stack is
about 10 times faster than that of Linux.

In this url, -http://lists.freebsd.org/pipermail/freebsd-net/2004-September/004840.html
- it is claimed that "FreeBSD can route 1Mpps on a 2.8GHz Xeon whilst
Linux can't do much more than 100kpps".

There are some that claim that BSD has a more consistent design, and
scale better than Linux. And then like the above, they say BSD is
lightyears apart from Linux in term of performance.

I don't know all that. I'm just a simple Linux user, but when people
start telling me stuffs like that, I got

.

Please don't treat this message as an attempt to incite flamewar
between the BSD and Linux camp, I am asking this question just to find
out the truth.

Can anyone help out, please ?

Thanks !

JEDIDIAH

2004-10-29, 3:58 am

["Followup-To:" header set to comp.os.linux.advocacy.]
On 2004-10-29, Penang <penang@myrealbox.com> wrote:
> The other day, on slashdot, someone mentioned that BSD's IP/Stack is
> about 10 times faster than that of Linux.
>
> In this url, -http://lists.freebsd.org/pipermail/freebsd-net/2004-September/004840.html
> - it is claimed that "FreeBSD can route 1Mpps on a 2.8GHz Xeon whilst
> Linux can't do much more than 100kpps".
>
> There are some that claim that BSD has a more consistent design, and

Where are the 2048 cpu SMP systems running BSD?

There are such Linux machines. They are sold by SGI.

[deletia]

--
Negligence will never equal intent, no matter how you
attempt to distort reality to do so. This is what separates |||
the real butchers from average Joes (or Fritzes) caught up in / | \
events not in their control.

Linønut

2004-10-29, 8:56 am

Penang poked his little head through the XP firewall and said:

> The other day, on slashdot, someone mentioned that BSD's IP/Stack is
> about 10 times faster than that of Linux.
>
> In this url, -http://lists.freebsd.org/pipermail/freebsd-net/2004-September/004840.html
> - it is claimed that "FreeBSD can route 1Mpps on a 2.8GHz Xeon whilst
> Linux can't do much more than 100kpps".

The presentation itself doesn't say that:

http://people.freebsd.org/~andre/Fr...-Networking.pdf

In any case, one would need to know more about the test, as well as the
configuration of the Linux kernel used for testing it.

Offhand, I wouldn't be surprised if FreeBSD TCP/IP were a bit faster than
Linux TCP/IP, just as Linux TCP/IP is faster than Windows TCP/IP.

By the way, are the guy who keeps trying to ssh into my box?

--
Penguins love icebergs.

Rex Ballard

2004-10-30, 3:56 pm

penang@myrealbox.com (Penang) wrote in message news:<f2375275.0410282108.425fe0a1@posting.google.com>...
> The other day, on slashdot, someone mentioned that BSD's IP/Stack is
> about 10 times faster than that of Linux.
>
> In this url, -http://lists.freebsd.org/pipermail/freebsd-net/2004-September/004840.html
> - it is claimed that "FreeBSD can route 1Mpps on a 2.8GHz Xeon whilst
> Linux can't do much more than 100kpps".

This is one of those "stacked benchmarks". FreeBSD is a much leaner
and lighter implementation, partly because it isn't as focused on SMP.
They didn't even try to include SMP until version 5. FreeBSD is a
very fast very reliable single processor system. The performance
enhancements are a good demonstration of the difference between UNIX
and Linux. Part of the performance difference is how memory is
managed and mapped for interprocess communications, particularly
between the TCP application and the intetd daemon. This includes
read-ahead and write-behind. Much of this was because high speed
routers were implemented using BSD. cisco routers, for example, use
BSD kernel and IO modules.

The big difference between Linux and BSD has more to do with the
nature of the licenses. The Linux kernel is copyrighted and licensed
under the GNU public license. The BSD kernel is copyrighted and
licensed under the BSD license. The GNU license has a mandatory
giveback clause. Essentially, if you enhance or fix the Linux kernel
you MUST send a patch and a description of that patch back to the
original author if you decide to publish a version of this patch.

The thinking here is that thousands of contributors have invested
time, money, resources, and talent to develop a product that is
desirable and viable. Those who have contributed to this "stone soup"
don't want a selfish company - such as Microsoft or SCO - to publish
this code, add a bunch of proprietary enhancements, drivers, and
features - then use these new features as leverage to bankrupt all of
the competitors who have contributed and played by the rules.

The BSD license has a different target audience and allows, even
encourages the development of proprietary derivative products.
Ironically, this has been one of the problems that plagues BSD code,
even today. For example, Microsoft uses BSD code for their TCP/IP
stack, of course SCO uses LOTS of BSD code in their offerings. Linux
also uses BSD applications. The problem of course, is that companies
who have banked on BSD code, such as BSDi and SCO have lost market
share to Microsoft (who is beating them with their own code), and
Linux (who includes BSD applications in Linux distributions). Keep in
mind that all of this is completely legal as a function of the BSD
license. The BSD license only requires that you include BSD and/or
the original author in the copyright notices for any derivative
products, that you indemnify the original author against any damages,
and that you do not use the BSD trademark in advertizing without
getting prior written permission (which has been given - usually
associated with modest donations).

The BSD license fosters lots of forking. There is FreeBSD, NetBSD,
BSDi, OpenBSD, and of course BSD code within AIX, Solaris, HP_UX, SCO
UNIX, Interactive UNIX, and any other SysVr4 or later version of UNIX.

All that said, there are still some standards that MOST versions of
*nix follow, and try to adhere to very closely. Linux doesn't
actually used BSD code, but tries very hard to use the same APIs, and
to adhere to all of the publicly available standards. Linux tries
very hard to make sure that code that compiles under GCC on other UNIX
systems will compile without modifications on Linux, and will function
in the manner documented.

The net result is that there is very little difference - at the
application and infrastructure level, between Linux and BSD. Each has
different file system structures and different kernel internals, but
they all run the same applications and can follow the same core
standards.

> There are some that claim that BSD has a more consistent design, and
> scale better than Linux. And then like the above, they say BSD is
> lightyears apart from Linux in term of performance.

This is actually a rather interesting observation, since there are so
many different direvative versions of BSD, including SysVr4 UNIX.
Most BSD based kernels are leaner and more streamlined, because they
don't try to handle things like massively parallel SMP and NUMA
environments, Clusters, and optimal interconnectivity between these
processors and numerous peripherals (multiple ethernet cards, multiple
drive controllers, multiple displays...). Linux also has a much
broader support for drivers through it's "modules" interfaces. This
is additional overhead, but reduces the need to recompile the kernel.

> I don't know all that. I'm just a simple Linux user, but when people
> start telling me stuffs like that, I got

.

Administrators would know in a few minutes which machine was Linux and
which was BSD, but the average user, walking up to the screen and
keyboard of either machine would have a hard time telling the
difference.

> Please don't treat this message as an attempt to incite flamewar
> between the BSD and Linux camp, I am asking this question just to find
> out the truth.

Bottom line, both BSD and Linux are excellent products. You will find
much more consistency between Linux distributions, since the primary
difference revolves around software installation and package
management, but both platforms are very good.

If you have a very small, older single-processor machine, that you
want to use as a standalone server, BSD is a very good choice. There
is much less need to upgrade, and if you have the drivers, you can
install BSD, stick it in a closet, and do all of the administration
through a remote interface - either vt100 style emulation (using ssh
or telnet), or using X11, or using VNC.

Both BSD and other UNIX implementations have been striving to be more
and more "Linux Compatible" because Linux has established an excellent
baseline which can be implemented by almost any vendor.

Ironically, even Microsoft has the ability to follow Linux
compatibility standards with Windows Services for Unix (SFU), which
adds about $100 to the already inflated price of Windows. Of course,
Microsoft has such a terrible track record for NOT adhering to the
standards, for adding proprietary extensions, and for breaking
compatibility with both UNIX servers and UNIX/Linux workstations, that
I would be very reluctant to trust SFU. Better to go with cygwin,
which is free, and can allow you to run many Linux applications on a
Windows workstation.

> Can anyone help out, please ?
>
> Thanks !

Shmuel (Seymour J.) Metz

2004-10-31, 3:56 am

begin In <f2375275.0410282108.425fe0a1@posting.google.com>, on
10/28/2004
at 10:08 PM, penang@myrealbox.com (Penang) said:

>There are some that claim that BSD

What is BSD?

>Please don't treat this message as an attempt to incite flamewar
>between the BSD and Linux camp,

How about among the Free, Net and Open camps? ;-)

Why not install one, two or all three and see how they stack up
against Linux in your environment?

--
Shmuel (Seymour J.) Metz, SysProg and JOAT <http://patriot.net/~shmuel>

Unsolicited bulk E-mail subject to legal action. I reserve the
right to publicly post or ridicule any abusive E-mail. Reply to
domain Patriot dot net user shmuel+news to contact me. Do not
reply to spamtrap@library.lspace.org

Paul D. Smith

2004-10-31, 3:56 pm

%% r.e.ballard@usa.net (Rex Ballard) writes:

rb> The GNU license has a mandatory giveback clause. Essentially, if
rb> you enhance or fix the Linux kernel you MUST send a patch and a
rb> description of that patch back to the original author if you
rb> decide to publish a version of this patch.

I don't want to get into license wars, and maybe the point is too
technical to matter much, but technically this is not what the GPL
requires. The GPL requires that if you modify GPL'd code you must make
available the full source of the derivative to anyone that you
distribute the binaries to, under the GPL license, for no more than a
reasonable cost of distribution. You _DON'T_ have to give it back to
the original author, unless you distribute it to the original author.

I mainly mention this because there _ARE_ licenses which require you to
send all changes back to the author, but the GPL is not one of them.

Cheers!

--
-------------------------------------------------------------------------------
Paul D. Smith <psmith@nortelnetworks.com> HASMAT--HA Software Mthds & Tools
"Please remain calm...I may be mad, but I am a professional." --Mad Scientist
-------------------------------------------------------------------------------
These are my opinions---Nortel Networks takes no responsibility for them.

Ralph

2004-10-31, 3:56 pm

Paul D. Smith wrote:

> under the GPL license, for no more than a
> reasonable cost of distribution.

As far as I know, you can charge as much as you want.

Eirik Seim

2004-10-31, 3:56 pm

On 30 Oct 2004 10:39:19 -0700, Rex Ballard wrote:

[ About FreeBSD and SMP ]

> They didn't even try to include SMP until version 5.

It may not have been much to look at at the time, but the SMP
branch was merged into 3.0-current in April 1997. What's new
in 5.x is SMPng.

--
New and exciting signature!

William Ahern

2004-11-21, 8:57 pm

Daniel Rudy <5n6o7.8d9c0r1u2d3y4.5s6p7a8m9@0e1m2a3i4l5.6p7a8c9b0e1l2l3.4i5n6v7a8l9i0d1.2n3e4t5> wrote:
> OpenBSD - The focus is security. Cryptographic support is built into
> the kernel. It will even encrypt the swapfile. If you are looking to
> build a NSA-DoD compliant machine, this is the one to use.

I don't think OpenBSD is compliant. Unfortunately, I believe compliance is
more a function of the number of acronyms supported, not proven security.
They want a million + 1 ways to create ACL's. Whether a hacker can subvert
them is another story entirely.

Fortunately, OpenBSD is the opposite. OpenBSD, like NetBSD, is primiarly
concerned with correctness of code and design, even if that means fine
tuning 30 year old code and systems. OpenBSD's emphasis happens to create a
robust system with regards to intrusion protection, be that code,
applications or processes.

Sponsored Links