Talk:Multiprotocol Label Switching

From Wikipedia, the free encyclopedia

Is MPLS a protocol? Should the category be changed to Network protocols? - Sridev 21:05, 16 Aug 2004 (UTC)

Well, yes and no. MPLS is more of a packet format, along with some simple processing rules for forwarding packets in that format. Whether that makes it a protocol is kind of a theological question. The MPLS spec is mostly there to allow people to build compatible hardware. You also need (in most cases) a setup protocol - which you may or may not consider part of MPLS (there are several different alternatives) - before packets can flow. Noel 07:57, 25 Oct 2004 (UTC)
I would call it an framework (defining the need) and an architecture (rules for writing protocols and expected behavior). Protocols flow between machines, so the expected internal behavior of a Label Switched Router (LSR) in forwarding packets [Note 1] and the expected behavior of a Label Edge Router in assigning packets to an MPLS path is not strictly a protocol
[Note 1] For all practical purposes, the generation of MPLS we are discussing here forwards packets. Generalized MPLS (GMPLS) has a framework and architecture for forwarding non-packet information, such as optical wavelengths (i.e., lambdas) in a Wavelength-division multiplexing system, time slots in time-division multiplexing such as SONET, and coordinates of the desired in and out ports on cross-connect switches (e.g., Digital cross connect system oe DACS). For all these non-packet forwarding modes, there still must be path setup protocols that interact with IP routing.Howard C. Berkowitz 15:56, 22 September 2007 (UTC)

Wire format + Semantics = Protocol. --Koshua 03:19, 7 Jan 2005 (UTC)

"Label Lookup and Label Switching [...] can take place directly into fabric and not CPU."

It would be nice to provide some explanation of this statment... why is this so? And re: a few paragraphs earlier, *why* was it so challenging ("impossible") "to forward IP packets entirely in hardware"? These are offered as a motivation for why this subject is important, but it's not explained (or linked) well enough to be understood. DKEdwards 00:31, 21 October 2006 (UTC)

I agree. Also, the link associated with the word "fabric" does not appear to work at the moment. Simeon Williamson 14:49, 9 November 2006 (UTC)

See Switched fabric and Forwarding Plane The basic idea is that the CPU just has to worry about setting up and tearing down paths, and doing fault circumvention for the paths, and the forwarding itself can be done with more purpose-built hardware Howard C. Berkowitz 15:56, 22 September 2007 (UTC)

MPLS is now also used in Ethernet Switching

Historical note that I thought was a little too detailed to go into the article - Tag Switching was originally Cisco's response to Ipsilon's high-speed switching product. (This was back from when everyone "knew" that switches could run faster than routers, and Ipsilon's clever marketing of their system [which used ATM switches] was proving vey effective.)

Another detail note: the issue with queueing delays and voice is that for voice to work, it needs to have low jitter (i.e. delay variance). You can get this in one of two ways: i) Have a big playback buffer, which allows you to smooth out the jitter, but which is (or was) expensive, and also increased the delay across the channel - and human conversational mechanisms tend not to work well with high-delay channels, or ii) Build a system with low delay and low jitter, for which you have to have short queueing delays, for which you have to have cells (a la ATM) on slow (and T1 was common when ATM was done) links if you intend to also carry large datagrams. Noel 07:57, 25 Oct 2004 (UTC)



In addition to MPLS, If I may suggest that a specific note be made on G-MPLS or Generalized MPLS. This is to basically to reflect the extension of MPLS into the time, wavelength, and fiber paradigm. The IETF common control and measurement plane (CCAMP) are working on standardizing the network plane across such multiple domains. http://www.ietf.org/html.charters/ccamp-charter.html

Contents

[edit] Suspicion of commercial unethical ad in external link section

I could not figure the interest of 3 differents links related to the same simulation product (NEST - QoSDesign) in the external link sections. At least I think negative impression should be balanced by providing links to Opnet and other broadly used simulation solutions.

[edit] Does the Experimental Bits section make sense?

I couldn't figure out that the Experimental Bits section made sense at all.

In the MPLS presentations I've seen, multiple tunnels between the same endpoints are used all the time - exactly because one can apply QoS differentiation to different tunnels. In fact, lots of the time, MPLS networks are sold as "this is the solution to the QoS problem". Delete section? --Alvestrand 08:23, 9 January 2006 (UTC)

Sad. It's documented here: [1]. --Alvestrand 15:51, 28 January 2006 (UTC)

The article lists the experimental bits as "QoS" bits. While Cisco (and likley other vendors too) use it for this purpose the RFC does not state a specific use for them. Should the article state they are QoS bits, or simply that its a common use of them? Theres nothing to say a vendor might use them for something else besides QoS. RFC 3032. - ML

I would take issue with the 'MPLS Deployment' and 'Competitor to MPLS' sections that refer to MPLS working in an IPv4 or IP-only environment as it is also supported in IS-IS environments.

I agree, MPLS doesnt care about L3 protocols when making forwarding decisions, it puts a 32 bit label in between the L2 and L3 headers. --Skydivemayday 03:01, 20 October 2006 (UTC)

Also, while ASICs have increased routing speed, they have done nothing to help with the fact that there are a variety of L2 transports deployed (ATM, Frame Relay, Ethernet, and SONET) which take a bit of work to interconnect every time you need to get from point 'A' to point 'B' and you would have to traverse these different types of L2 environments. MPLS adds an 'abstraction' layer which, when coupled with the VPN capability, reduces the amount of provisioning required.

If any comparisons to L2TPv3 are to be made, it should also be noted that MPLS supports both point-to-point and multipoint VPNs and L2TPv3 truly is IP-only and point-to-point. Joseph chapman 04:09, 24 April 2006 (UTC)


[edit] QoS topics missing

MPLS is everywhere described as being the ultimate solution to QoS problems by supporting Integrated Services (IntServ) and Differentiated Services (DiffServ) at the same time. It can also be used in traffic engineering for things like load balancing on ip-networks (because it's not using shortest path routing like ip always does) and defining backup paths in the case of network failure.

I think these things should be added because for a lot of internet service providers and other communication companies these are the main reasons for using mpls and, of course, as a replacement for atm. A lot of telco companies are now offering voice over ip and tv over ip so for them mpls is a way for ensuring QoS (which is of great importance for realtime services).

06:46, 01 October 2006 (UTC)

[edit] MPLS with non-IP protocols - clarification?

The first paragraph states that MPLS can be used to carry native ATM, SONET, and Ethernet frames. This contrasts with a late section which states MPLS cannot be compared to IP as a separate entity because it works in conjunction with IP and IP's IGP routing protocols. Could someone add a little more detail on the first point? In particular I find it puzzling that the MPLS header diagram does not show any indication of packet length, and so presumably depends on the IP packet length indicator. How does this work with non-IP content? Scottwh 13:33, 12 December 2006 (UTC)

[edit] MPLS vs Ethernet

There's a big debate about this (MPLS vs PBT Ethernet) at the moment. Can someone please add some OBJECTIVE, NON-RELIGIOUS, SCIENTIFIC info in about the main differences between the two formats? But are these protocols comparable with each other —The preceding unsigned comment was added by 61.222.209.68 (talk) 09:48, 2 April 2007 (UTC).

"Remember that a LER is not usually the one that pops the label. " is the first mention of popping. How can you remember anything? Perhaps a previous edit has removed the relevant pre-explanation?

[edit] IEEE 1355

IEEE 1355 is a completely unrelated technology that does something similar in hardware.

Sounds like a "deafening silence" aka Oxymoron. Can someone explain the sentence for non-native speakers? --80.136.108.252 (talk) 16:04, 5 April 2008 (UTC)