Skip site navigation (1)Skip section navigation (2) 
Date:      Wed, 15 Apr 1998 11:35:37 +0700
From:      Charlie & <root@nnd.itfs.nsk.su>
To:        hackers@FreeBSD.ORG
Subject:   CFD: 'cce' - "concatenated" ethernet driver
Message-ID:  <19980415113537.01993@nnd.itfs.nsk.su>
next in thread | raw e-mail | index | archive | help 
	What do you think about such an interface ?

	Brief description of 'cce':

   It is an interface (such as ed0, fx0, sl0, ppp0) which have some
number of "phisical" ethernet interfaces (PEIs).

   As an interface it can be 'ifconfig'ed to have some IP address
from some (sub)network (and PEIs are 'unnnumbered').

   It then 'received' all the packets sended to this address through
all of the PEIs.

   And I can see (at least) three "modes" of 'send' operations
with respect to the PEIs:

	- "HUB (REPEATER ?) mode" - all the packets duplicated to
	  all of PEIs;

	- "BRIDGE (or is it SWITCH ?) mode" - the packets to
	  destination IP address is sended only to PEI which has
	  the host with this IP adress connected to
	  (resolved through ARP potocol);

	- "PARALLEL mode" - packets are sended to PEI in turn
	  to 'load balance' them or to 'double throughput'.

   In first two cases other hosts in network are coneected to the
one of PEI (and use it as usual 'ordinary' ethernet). And in last
case other host(s) also use 'cce' in PARALLEL mode.

   ARP for such an interface needs some modification depending of
mode of operation but primarily will be based on the 'concatenation'
or 'combination' of ARP requests and results for/from PEIs.


   If the 'cce' driver will have small overhead then it can be
installed in (one of) the  office's FreeBSD server(s) and
(without splitting IP adress space which requires regular
renumbering or wasting of IP adresses):

	- increase the phisical lengh of the ethernet network
	  (REPEATER mode);

	- decrease the collision rate and increase the usefullnes
	  of the highly populated network which can be devided
	  to 'highly connected' subnetworks (SWITCH mode);

	- increase throughput or establish some kind of
	  "fault tolerant" network between critical servers in the
	  network without switching to the higher speed/price
	  hardware (PARALLEL mode).


	May be there already is some implementation of such
interface(s) ? (Under other name I hope :-).

	Or are there some volunteers ?

	Can some of 'guru's estimate man-month price of such
implementation ?

	N.Dudorov



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message

Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19980415113537.01993>