[84623] in North American Network Operators' Group
Re: image stream routers
daemon@ATHENA.MIT.EDU (Lincoln Dale)
Sat Sep 17 21:34:15 2005
Date: Sun, 18 Sep 2005 11:29:52 +1000
From: Lincoln Dale <ltd@interlink.com.au>
To: Dan Hollis <goemon@anime.net>
Cc: "Edward B. Dreger" <eddy+public+spam@noc.everquick.net>,
"'nanog@merit.edu'" <nanog@merit.edu>
In-Reply-To: <Pine.LNX.4.63.0509171753210.21595@sasami.anime.net>
Errors-To: owner-nanog@merit.edu
Dan Hollis wrote:
>> right. what i'm pointing out is that if Imagestream routers really
>> ARE capable of >OC12 (and perhaps multiple of them) then its unlikely
>> its s/w-based forwarding.
>
> doesnt mean they are violating GPL to do it. look at nvidia for example.
agree. all i'm saying is that it is "unlikely".
OC12 at minimum-packet-size is a little over 3.2M PPS unidirectional
traffic. bidirectional its 6.4M PPS.
it is very unlikely that a non-modified linux kernel can do anywhere
near that. its also incredibly unlikely that even a modified kernel
could do that.
the only possible scenario where i can see unmodified linux doing that
is if you used "Fast Routing" which essentially is DMAing from one NIC
to another, removing any ability to do any fancy queueing, no ability to
do ACLs or any form of traffic accounting.
in terms of 'router characterization', i doubt most folks would want
such a thing on either a core-router, peering-router, transit-router or
aggregation router.
so: i'd say its far more likely that its h/w-based forwarding, perhaps
FPGA-based.
cheers,
lincoln.
