[2213] in linux-net channel archive

home help back first fref pref prev next nref lref last post

Re[2]: IP forwarding bug in 1.3.73

daemon@ATHENA.MIT.EDU (Paul McCallum)
Fri Mar 22 13:47:21 1996

Date: 	Fri, 22 Mar 1996 12:19:50 -0600
From: Paul_McCallum@crow.bmc.com (Paul McCallum)
To: linux-net@vger.rutgers.edu

     I am just learning about TCP/IP so I dont understand all its re-send 
     implications at this time but these are my recollections for TU sizes on 
     phones lines:
      
     1) small sizes reduce the time to detect and re-transmit bad frames on 
     low-quality lines.  modern facilities have reduced this need to a large 
     extent.
     2) small sizes make for better line sharing.  PPP by definition cannot 
     share a line.
     3) small sizes are required by some hardware due to cost.  again time has 
     cured this not to mention the fact that both devices are programmed and 
     memory is controllable.
     
     The bottom line is I agree, MTU should be reasonably large, less management 
     overhead is better.


______________________________ Reply Separator _________________________________
Subject: Re: IP forwarding bug in 1.3.73
Author:  Todd Graham Lewis <tlewis@mindspring.com> at UNIXLINK
Date:    3/20/96 10:13 PM


On Thu, 14 Mar 1996, Zoltan Hidvegi wrote:
> We have a ppp dial-in server running linux-1.3.73.  Normally it uses 296 
> MRU/MTU on the ppp interfaces and 1500 MTU on the ethernet card.
> Unfortunately it cannot forward UDP packages larger that 296 bytes correctly. 
> 
> When I raise the MTU on the ppp line to 1500 everything works fine.
     
Why does everyone insist on running such a small mtu on their ppp 
connections?  Larger mtu's are tougher to tweak, but everyone should be 
aiming at the largest mtu they can handle, and rarely do you really have 
to bump it below 1000.  Smaller mtu's will cause excessive fragmentation, 
with end-to-end mtu discovery, they place a much larger load on the 
intervening routers, which almost always includes the backbone routers, 
which are _way_ overburdened as it is, and it decreases performance.
     
I simply do not understand why people insist on doing this.  If there is 
a reason, could someone please clue me in to the big secret?
     
Todd
     
____________________________________________________________________________ 
|                                                                          | 
|Todd Graham Lewis      Core Engineering       Mindspring Enterprises, Inc.| 
|tlewis@mindspring.com                                       (800) 719 4664| 
|__________________________________________________________________________|
     
     
     


home help back first fref pref prev next nref lref last post