Re: Broadcast Packets Evil?

On Tue, 23 May 2006, in the Usenet newsgroup comp.os.linux.networking, in
article <slrne764uj.cdc.spamtrap@xxxxxxxxxxxxxxxxxxxxx>, digitlcoup wrote:

On 2006-05-23, Moe Trin <ibuprofin@xxxxxxxxxxxxxxxxxxxxxx> wrote:

To a minor extent, it depends on the network topology. If you have 10
hosts on a switch, the bandwidth of the switch should sustain the connection,
and they'll just be a momentary spike in network traffic. If you have an
unswitched network, or it is large, then people start stepping on each other,
backing off, stepping on each other, backing off - Lather, Rinse, Repeat.

That's why a large network should be setting up VLANs to prevent such
problems, but I'm sure you anticipated this response.

See if you can find a copy of Robert Hart's old 'Linux IP Subnetworking'
mini-HOWTO from the mid-1990s or the retitled 'IP Sub-Networking Mini-Howto'
last revised in August 2001 which merely corrected a minor (obvious) typo.
I'm not seeing a copy on sunsite any more, but google may find it for you.
It is obsolete, talking about classful networks (Class A, B, C) which went
the way of the do-do in 1993 with CIDR (RFC1519), but the concepts are good.

In the past, IANA was handing out IP blocks like they were in-exhaustible,
and it was possible for a large organization to get a 'Class "A"' (/8), or
'Class "B"' (/16) assignment. While a large block might be appropriate to
a backbone/major provider, some of the blocks were allocated to ordinary
companies. I can assure you that NO ONE is using those blocks as one
massive network. The original Ethernet (10Base5 a.k.a 'thicknet') had a
cable length limit of 500 meters (1640 feet), and allowed transceivers
(attachments to the cable) every 2.5 meters (8.2 feet). (Later, at least
one manufacturer built multi-port transceivers - one vampire connection
split to 4 AUI connectors, but they were not very common.) There could be
a maximum of two repeaters (that bridged between cable segments) between
any two systems. Assume a single main cable, with 200 drops attaching to
200 cables with 199 hosts (and one repeater) on each, you could have a
maximum of 39,800 hosts on one massive (but not very useful) network. Maybe
that's not such a good idea. The solution has been around for a long time,
and is called 'subnetting'. See RFC0950, which dates from August 1985. If
you are interested in earlier work, see RFCs 0917, 0925, 0932, 0936, and
0940. The _concept_ of subnetting is a lot older than that, going back into
the 1970s.

Where I work (sorry, NDA), our networks were originally set for /22 masks
or 255.255.252.0, allowing 1022 hosts on a given wire. In the mid-1990s,
we started altering the physical layout, breaking the 10Base5 cable into
segments with no more than 70 hosts on a given segment, and these segments
being connected using an early design of the (now common) switch. We also
put servers onto their own switch ports, to further isolate the traffic.
The results were gratifying. Subsequent redesigns of the topography using
first 100BaseT and now Gigabit (both copper and fiber) and a few more file
severs has further improved the traffic flow. The 'broadcast' domain
remains at 1022 (possible) hosts, but as we've disabled ping responders
on workstations and some servers, the only common broadcasts are ARP
packets which are a drop in the bucket.

Old guy
.

Relevant Pages

  • Strange results from a tcpdump, can anyone help?
    ... traffic was going ballistic on most ports in the network. ... other hosts went to normal (i.e. the only traffic you could see were ... packets from the same vlan destined to other hosts outside ... If it was simply a bad switch with a bad port that had lost it's mac ...
    (comp.dcom.lans.ethernet)
  • Re: Strange results from a tcpdump, can anyone help?
    ... traffic was going ballistic on most ports in the network. ... other hosts went to normal (i.e. the only traffic you could see were ... packets from the same vlan destined to other hosts outside ... If it was simply a bad switch with a bad port that had lost it's mac ...
    (comp.dcom.lans.ethernet)
  • Re: Multihoming Windows 2000
    ... They are looking to connect all hosts to the core network via 1 NIC. ... will connect to another switch and this network is considered the ...
    (microsoft.public.win2000.networking)
  • Re: RDP Connections - Freeze or Reconnect
    ... comes from the switch and then connects into what I can only describe ... as an extension cable with a female/female connector. ... The server is one of the most critical parts of the network and as such it should have a professional high-quality cable connection to the switch. ... You should consider bringing in your own switch as well as a few long network cables for testing. ...
    (microsoft.public.windows.terminal_services)
  • Re: setting up a network
    ... I have a switch and cables etc. So far I have only been ... What is the purpose of the network, and what exactly do you mean by "hook my ... you need a router to use more than one computer. ... You then use regular patch cables to connect the computers ...
    (microsoft.public.windowsxp.help_and_support)