What is the difference between up and down link on LAN switch?

Followup to: By author: "Henry Kiefer" In newsgroup: comp.arch.embedded

The only difference is which way the pairs are wired. Also, for gigabit switches and some newer 100 Mbit switches there is no difference at all; they work either way.

So yes, the difference is *exactly* the same as using a crossover cable.

This is the modern equivalent of the old DTE/DCE confusion in the RS-232 standard; it makes sense when you know what is a DTE (terminal) and what is a DCE (modem), but gets really confusing when people try to be clever about it (I have an embedded system wired as a DCE, for example.) Routers, which includes cable modems, are functionally hosts, equalling computers (DTEs), and hubs/switches are the equivalent of a DCE. However, since you need to be able to connect two hubs/switches together you have to switch one of the ports to "uplink" mode, which makes that port wired as a host.

-hpa

H. Peter Anvin schrieb in Nachricht ...

Thank you H. Peter for your lengthly explanation! How many switches can be cascaded? What is the time delay between ports of a switch?

I know the problem with DCE-DTE variations on RS-232 (I worked approx. 20 years with embedded systems).

My problem at the moment is, that my two PCs don't work with the Motorola cable-modem at the SAME time. Only after resetting the modem the first come-up PC gets the internet connection. Motorola says, up two 32 PCs can be running with the integrated DHCP server. My provider KabelBW says, connecting a couple of PCs is allowed and should work.

cu - Henry

It's just crossing the TX/RX pair. An 'uplink' port is like a network card port and can be plugged into the same things. The other 'down' ports have their pairs swapped so they can be connected with straight-through cables to network cards.

For ambiguous devices like a cable modem just try both and see how it's wired.

There are such things on managed switches (called trunk ports) but it has nothing to do with the physical wiring.

Initially it doesn't. It broadcasts packets (to all ports) destined for MACs that it doesn't recognize. When the target responds the switch sees what port it was on and remembers it for a while. If you power cycle your switch and then watch the activity lights you can see it happen.

Yes.

Yes.

No. Not for your average consumer-grade switch.

They have a cache that keeps track of which MAC addresses are sending packets in which ports. The switch assumes that packets _to_ MAC address aa:bb:cc:dd:ee:ff should be sent out the port where packets _from_ MAC address aa:bb:cc:dd:ee:ff come in.

If the switch needs to send a packet to an MAC address from which it hasn't seen any packets, then it just sends it out all the ports.

As many as you want. There is a limit to the number of hubs that can be connected in a single path, but switches create an isolated collision domain on every port, so there is no limit.

Not much. It depends on the switch, but I would expect it to be not much more than the packet length plus a few hundred microseconds of overhead.

You need NAT, not DHCP. You should get a 'cable router', which is a little switch with integrated NAT. Otherwise I'd expect the behavior you're seeing.

first one is crossover

Pozdrawiam.

-- RusH //

Well, there are some limits that might bite...

Normal spanning tree only supports up to 7 hops. You can usually adjust this max slightly, and non-stp participant bridges and hubs don't count as hops, but it's still a limit of some importance in a large network.

Other bridge-layer protocols might impose similar limitations. The only other one I recall that propogated beyond a single link is GARP, and I don't remember any hop limitations; no doubt it runs atop the spanning tree to avoid the issue.

There is also the scalability issue of having many hosts in a single broadcast domain. Depending on the protocols in use, some smaller hosts (print servers, network cameras, etc) can be measurably affected by the overhead of receiving lots of broadcast "noise".

The broadcast problem is slightly worse if you make heavy use of IP multicast; in large installations with lots of multicasting there are little traffic "leaks" in all the various elements from nics to routers that add a bit of extra work all around.

cu - Henry

OK!

Ben Jackson schrieb in Nachricht ...

be

seeing. Hm. I will see what the provider says. Is it possible to use a switch with integrated NAT and leave the cabel-modem think there ist only one PC connected? How is the "switch with NAT" called, so I can buy one?

cu - Henry

OK Grant - I think your're working on really great networks of hundreds of stations or more. Hopefully I don't have so much children chatting all the day ;-)

So long - Henry

Usually described as a cable/broadband router. Linksys do the BEFSR41 (There's also the WRT54G with wireless). Netgear have one as well (RP614).

Pozdrawiam.

OK RuSH! Thanks and dobredanje *lol

- Henry