Tuning Gigabit

David Gilbert dgilbert at velocet.ca
Sun Jun 29 13:38:09 PDT 2003

>>>>> "Gregory" == Gregory Sutter <gsutter at zer0.org> writes:

Gregory> The submission deadline for this year's BSDcon is long past,
Gregory> but you can always finish it up for next year's USENIX or
Gregory> BSDcon.  Alternatively, you could write up your results for
Gregory> publication in the Daemon News ezine
Gregory> (<http://ezine.daemonnews.org/>), a monthly publication since
Gregory> 1998.  We accept all types of BSD and Unix-related articles
Gregory> from short newbie articles to journal-quality pieces.  To
Gregory> submit work or ask questions, email
Gregory> <articles at daemonnews.org>.

All worth mentioning, I admit ... which is to say I would consider
it... but from a purely academic point of view, having a properly
accredited journal (with an appropriate editorial board and a proper
peer review process) would do more for both *BSD and the individual.

Certainly, I keep abrest of developments in our little world... and I
respect some of the work that is published in various places for
FreeBSD, but in the end, my goal is to have FreeBSD stack up against
the Cisco and Juniper routers ... and that requires a broader audience
... both for criticism and for exposure.

My basic argument on generic OS routing is this: If you never do
anything unexpected, a closed system router (not even just closed
source) will do you fine.  They're generally reliable beasts and the
facts about them are known to thier own liberate.

If you need something to pass 1 gigabit of traffic with a maximum of
600 kpps, a CCNE should be able to tell you which Cisco to select
... and you shouldn't have any problems until your needs substantially

No such advice exists for FreeBSD or Linux ... or indeed Win-XP.
While the latter two don't interest me, it should be no less
possible.  Can FreeBSD pass 600 kpps?  Can FreeBSD pass 1 gigabit?
It's taken months, but I can say yes ... with the right hardware.

Now... we have an uphill battle.  Common knowledge is that the only
way to massivly route packets are custom ASIC's.  Little mention is
made of route clustering and other technologies that would scale
generic system routing.

Even though it's reasonably well known that the modern desktop
processor has become rediculously powerful.


|David Gilbert, Velocet Communications.       | Two things can only be     |
|Mail:       dgilbert at velocet.net             |  equal if and only if they |
|http://daveg.ca                              |   are precisely opposite.  |

More information about the freebsd-performance mailing list