Linux vs FreeBSD, and other *NIXes

[peterman at eecs.tufts.edu (Charles Peterman)]

Let me comment as someone who has done a fair amount of I/O testing
and admin work between the two.

Comparing FreeBSD 4.x stream to RH Linux 2.2 and 2.4 kernels.
I am shooting from off of the cuff, so I do not have the hard numbers
with me right now.

The freeBSD filesystem UFS, was less prone to data loss than
ext2fs on Linux. Sudden system outages did not cause the UFS as
much grief as ext2fs.  This was true early on in the 4.x stream
with soft updates turned off, and is with the current release (4.6)
with soft updates turned on.  Reiserfs performed the best for Linux.

In terms of network IO, given roughly equivalent system loads FreeBSD was
more reliable.  In that tcpdump -w foo.txt -i whateverdev, would not drop
packets on FreeBSD, whereas under Linux it would consistently drop between
3-10% of the packets. This occurred with a small variety of 10/100 network
cards writing to a variety of devices.

When stable device drivers are required for Linux, they are often lifted
from FreeBSD.  See the whole stink from earlier this year over Soren
Schmidt's ata driver magically appearing in the Linux kernel without
his copyright or even a note of thanks.  I prefer to run code that is
being executed in the OS it was written for, where the developers have a
clue as to why they wrote it that way.  Cut and paste kernel coding scares
me.  Often with Linux, that is what you get.

As for administration, I find it much easier to babysit a freebsd box.
The installation allows you to tie it down fairly tightly, though not
as tight as OpenBSD. After getting arsed by RH's insecure font server on a
default install of  7.1 (or 2?), I don't think security when I think RH.
Although I understand Debian is a little less foolish and SecureLinux go
a ways to make this better.

I prefer the freebsd pkg system to the others, but that is mostly due
to its centralized control and ability to cleanly resolve dependencies.
There are other features that might make a linux package manager more
appealing to you. If I am feeling paranoid, I can install from ports
instead of packages, and have everything compiled locally.  Either way
is convenient, and results in the same binaries most of the
time.

The downside of FreeBSD is that you don't always get the latest stable
software.  Xfree86 4.2 was just made a part of the default installation
as of FreeBSD 4.6.  Security trumps features, so software will often
languish on older versions until a compelling security hole/fix
necessitates upgrading.  You can roll your own newer
version of SpiffyServer6.x from source if you like, but chances are it
won't be available in either ports or packages.  Since the source is
mostly the same, security fixes in FreeBSD and Linux become available at
roughly the same time.


Some people who are doing professional sysadminning are still running
Linux 2.2 kernels, attempting to avoid the 2.4 morass. In general, you
see less of a clusterf*** in FreeBSD.  If it wasn't for it fabulous
uptime,broken only by power outages, I would probably upgrade one
of my firewalls from 4.2 to 4.6.

Take this with a grain of salt, I am not in the guru category of
sysadmin, my command lines only have two pipes, at the most.


-Charles

On Sat, 22 Jun 2002, David Kramer wrote:

> So does anyone have *recent* facts or anecdotes about the technical merits
> of one over the other?
>
> Historically, FreeBSD has had a more secure and debugged TCP/IP stack, but
> I haven't heard that claim in a while.
>
> This is really a scaled-down "cathedral and bazaar" experiment, as linux
> is developed by thousands with very little official oversite, while BSD is
> guarded by a smaller group.
>
>
>
>
> In line with some other discussions on this list. from a
> performance/scalability/reliability perspective, the areas that
> differentiate *NIXes the most are:
> - Scheduling
> - Memory management
> - Filesystem scheme (when to write, how much to read, etc)
>
> One of the reasons Sun did so well in the past is that they had excellent
> scheduling schemes.  In fact, it was replaceable, so you could change the
> scheduling scheme to suit your purposes.  As I understand it, Linux's
> scheduler is one of it's weaknesses, and a reason it benefits so much from
> multiprocessors.  Dunno about FreeBSD.
>
> AIX played freakish games with memory management.  The upside is that
> memory management is very efficient and usually sucessful.  The downside
> is that memory is constantly borrowed from disk cache to heap to shared
> libraries so it's hard to know exactly how much memory is actually free.
>
> -------------------------------------------------------------------
> DDDD   David Kramer                           http://thekramers.net
> DK KD  "That venture capitalists are willing to take any level of
> DKK D  risk, even a modest one, after all that has happened in the
> DK KD  ecommerce sector, is inspiring.  They might almost be
> DDDD   capable of becoming Red Sox fans"               -Keith Regan
>
> _______________________________________________
> Discuss mailing list
> Discuss at blu.org
> http://www.blu.org/mailman/listinfo/discuss
>