Talk:Cut-through switching

From Wikipedia, the free encyclopedia

[edit] Material cut in Sept 2004

I've removed the following material from the article. This seems to be a POV diatribe about ATM switching although it has some useful material and should be preserved. Cut through switching can exist in a pure ethernet context and in that case requres only one switch to deliver benefit; this means that this material is wrong in this context. Mozzerati 20:17, 2004 Sep 25 (UTC)

In standard routed networks, each node looks up the destination address and forwards a packet according to the directions in its internal routing table. With cut-through routing, a switched network core is embedded in a router network--picture it as two concentric circles with the switch core in the middle. At the edge of the switched core, an address-resolution process gives the incoming edge node the address of the edge node closest to the destination. A virtual circuit is then set up between them, which "cuts through" any other route examination along the way.

ethernet switches don't need to do this to take advantage of cut through.. needs to be cleaned up to say ATM

The primary benefit of this technology is improved performance of applications across the network, which is achieved by reducing delays. But it may not be easy to realize.

it's very easy to realise in ethernet (one interface command [1] on an IOS router)

It takes at least four consecutive cut-through-aware interfaces to make routing work. You need a device in the core that's cut-through-capable (such as an ATM switch) with an incoming and outgoing interface, and you need an edge device at each end of the core's pair of interfaces. Without all this, you can't set up the connection.

not in the case of IP into ethernet or other similar stuff

However, as the size of the core network decreases, the benefits of cut-through routing may become insignificant. Vendors are now building "super-routers" with forwarding rates of hundreds of thousands of packets per second. Speeding up a pesky chunk of congested network with one of these devices might be easier (and cheaper) than deploying cut-through routing and its associated switches. A fast router is a brute-force solution, but sometimes a hammer loosens the nut better than a wrench.

why not do both (use a super router with cut through)?

For incremental network changes, cut-through routing's benefits are largely affected by the size of the network portion that can be adapted for cut-through routers. A little dab of ATM here and there won't do much except prepare you to deploy more ATM. If that's the only benefit of an ATM-based cut-through-routing solution, it may be insufficient to justify any significant investment in the approach. Before you consider switched backbone networks using cut-through routing, make sure you'll have enough of it to improve performance.

sounds like anti ATM propaganda (don't get me wrong, I hate ATM just as much as the next guy, but this should be encyclopaedic.

It also follows that the more devices you can cut out with your cut-through connection, the better. This means that users planning to adopt cut-through routing should also plan to deploy switches (again, ATM switches are an example) broadly in their networks. Doing so will bring the core as close to each traffic source and destination as possible and reduce the number of network elements that have to handle the data in traditional router mode.

latency is addative, not dependent on element. This means that every stage where latency is reduced helps.

[edit] Current state

The article is a little bit confusing. The cut-through switching is implemented at the Level 2 (Data Link) from the OSI Reference Model, where IP does't exists, there are only Ethernet/IEEE 802.3 frames. Maybe that's where the confusion is from: IPv4's Checksum field applies only to the header, but the Ethernet's FCS (Frame Check Sequence) applies to all the frame, that is, including the data.

Also, the possibility for a cut-through switch to forward an incorrect frame exists. That's its main drawback, but it's really fast. —Preceding unsigned comment added by Marius p (talk • contribs)