Date: Tue, 10 Oct 2000 22:32:39 +0200 From: "Siegbert Baude" <Siegbert.Baude@gmx.de> To: "Alfred Perlstein" <bright@wintelcom.net>, "Mike Meyer" <mwm@mired.org> Cc: <questions@freebsd.org> Subject: Some basics about switches, hubs, full-duplex, half-duplex and OSI (was: Re: Network trickles) Message-ID: <00b101c032f9$3c054e00$4011a8c0@wohnheim.uniulm.de> References: <8049746@toto.iv> <14819.17289.859838.366697@guru.mired.org> <20001010093543.R272@fw.wintelcom.net> <14819.17993.590268.201193@guru.mired.org> <20001010095834.T272@fw.wintelcom.net>
next in thread | previous in thread | raw e-mail | index | archive | help
> If you have a hub then you'll want to toggle it to half-duplex, if > you have a switch, most likely full duplex. > Btw, cross-over cables can function in full duplex. > The most switches can auto-negotiate so that if you're set up > incorrectly you should be ok, however the real killer is trying to > use full-duplex on a hub, which never seems to work. > -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] Alfred is right in all points referring to practical handling of hubs and switches. As I just have some time (this really happens now and then :-) ), I try to shed some light on the terms and facts (buying a book or reading a good magazine like my favourite cīt here in Germany, nevertheless is highly recommended.) To whom it may be useful: MAC: Media (physical cable) Access Control, implemented in the network controller itself (e.g. your NIC). Each controller is defined by a world-wide unique number (MAC-address). Logical network addresses (e.g. IP-addresses) are always mapped to MAC-addresses for actual data exchange. NIC:. Network Interface Card. The thing you plug into your computer to get access to a network, (nowadays normally Ethernet). Broadcast: Typical traffic on Windows-networks ;-) They look for hardware addresses of other network controllers in the same net. Half-Duplex: Only one device is sending packets over the cable at the same time. Method to achieve this is CSMA/CD (Carrier Sense Multiple Access/Collision Detection). The first means, that every device is listening to the cable and sending packets only, if it is unused in the moment. It is therefore possible, that two devices are accidentally sending packets at the same time (so-called collision). The packets are lost and will be resent after a small delay. Full-Duplex: Two devices are sending packets over the cable at the same time as up- and downstream. There is NO collision detection in full duplex mode! Only possible therefore with (real or virtual) point-to-point connections (two NICs with cross-connect cable or switch with standard cables). Repeater: Amplifier to extend the network physically over the maximum possible cable length. Fully transparent (invisible)for the network. Hub: A multi-port repeater acting as cable concentrator for a star topology. Hubs and Repeaters connnect devices to a "collision domain". All collisions are seen on the whole net. To avoid this you need a bridge. Bridge: Separating a net into subnets (the subnets can have different sorts of cabling, e.g. coax and twisted-pair). Packets with sender and receiver in the same subnet arenīt transported to the other subnets. As you must look into the packets for the destination MAC-address to achieve this, this can slow down your traffic a little bit. On the other hand you increase the bandwidth within the subnets, as only a part of the packets are travelling around. Normally a bridge can handle only one connection at a time. Switch: Including the functionality of a hub, a switch is also a multi-port bridge which offers logical point-to-point connections between the attached devices (Full-duplex is therefore possible with switches!). This is achieved by looking for the MAC-address in the packets and transporting them via an internal high-speed bus (called backplane) to the port only, where the device with the receiverīs MAC-address is attached. This is done for all connections simultaneously, so to achieve full network speed the backplane must have a bandwidth of (possible full-duplex) 2 X (number of ports) X (bandwidth of the port). For a 10-port 100Mbit switch: 2Gbit/s ! All devices up to here transport broadcasts. To prevent this you will need a router. Router: A device connecting two network segments logically (a repeater does this physically). Using network layer information and routing tables, a router transfers packets, which arenīt for the local network segment to the next segment (or next router, if the destination segment isnīt reachable directly). Gateway: In earlier times routers were called gateways. According to nowadays use of the term "gateway", it passes data between networks having similar functions but dissimilar implementations. A router therefore is an OSI layer 3 (network layer) gateway, and a mail gateway is an OSI layer 7 (application layer) gateway. OSI: Open Systems Interconnection. A collection of protocols, developed by ISO. OSI Reference Model: This structure describes a computer network with a seven-layer architecture. Each higher level is working on top of the others. Layer 1: Physical (bit transport) level. Defines mechanical and electrical interfaces, e.g. the cables. Ethernet and Token Ring are on this layer. Hubs and Repeaters work here (they are too dumb to know anything about the bits they are transferring :-) ) Layer 2: Data link level. Transport is secured here by the MAC (therefore often called MAC-layer). Bridges and Switches work on this layer, i.e. they must know how to extract MAC-addresses and therefore arenīt transparent to the LAN-standards, theyīre working on! Layer 3: Network level. Independent of media and topology this layer is for the real transport of data over information over networks. IP is based on this layer. Routers therefore are working on layer 3. Layer 4: Transport level: Interferes between the top three levels (application system) and the bootm three levels (transport system). Makes transport reliable and offers the possibility to transport different services on the same transport mechanisms. TCP and UDP are here. Layer 5: Session level: Synchronization and flux managment Layer 6: Presentation level: Connection setup and termination, also presentation of data. Layer 7: Application level: Communication between applications. So the 5 layers of the protocol family often abbreviated with TCP/IP is OSI layer 3-7. If you read about "Layer 3 Switches", you will see, that such a thing doesnīt exist. Itīs a marketing term for switches with the additional features of a router. Nice to know: Switches differ in the bandwidth of the backplane, but also in the number of MAC-addresses they can use. Not important for home use, but if you attach large cascades of hubs and switches, a port must know all MAC addresses of devices wired to it. This can be a large number. If you need the support of SNMP in your switch, you have to decide by yourself :-) If everything is set to auto-negotiation, things work normally. Problems arise, if something is adjusted manually. The definitions arenīt fully clear, if this implies, that everything has to be setup manually then. So there are different implementations, what to do in this case, which often leads to misunderstandings between the devices. Hope this helps a little bit Siegbert To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00b101c032f9$3c054e00$4011a8c0>