[lkml]   [1996]   [Jun]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: A Humble Suggestion for 2.0/2.1
> Dennis of etinc is a BSD bigot, and that's ok.  Good or bad, BSD does 
> twice the packets per second that Linux does, and it is a more logical
> choice for a PC-based router. (This is not to slight any of the

Slight problem.

1. Dennis is not a BSD bigot, he's someone giving advice he thinks is right
I also when asked by people "Can I run BGP4 routing on Linux" answer
"No get an XXX" - I tend to suggest a dedicated router however 8)

2. Except with very large routing tables (2,000+ routes) Linux 2.0 CREAMS
the BSD stack. We are getting TCP latencies on 100baseT surpassing
ultrasparc machines with just a Pentium 133. (*)

(*) And thats Larry McVoy's benchmark not a Linux hackers benchmark

> networking crew for Linux; Van Jacobson is a hard act to follow, and no
> one expects that we will surpass BSD overnight. Everyone expects that
> we will eventually, but that's a different post.)

This one is really funny. About 2/3rd of the 2.0 speedup comes from following
ideas like copy/checksum in one pass. Originator: Van Jacobson, BSD folks havent
got around to adding it yet. Our base sk_buff scheme is very close to Van's
pbuf work (linear buffers again). We actual probably use more of Van's tricks
than 4BSD - and yes Van is VERY VERY good. We've also done a few other things
the BSD folk still lack - accurate ack timing, sneaky fast retransmit tricks
with congestion window sizes, and properly generated (ie nearly secure)
tcp sequence numbering.

The pre2.1 code adds about the last major trick VJ invented - the famous
sparc 30 instruction tcp fast path.


 \ /
  Last update: 2005-03-22 13:37    [W:0.046 / U:0.564 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site