Talk:Maximum transmission unit/Archive 1

Requested move

MTU (networking) -> maximum transmission unit -- expanding acronym will avoid need for ugly bracketed disambiguation Plugwash 16:56, 23 April 2006 (UTC)

Survey

Add *Support or *Oppose followed by an optional one-sentence explanation, then sign your opinion with ~~~~

• Support per nom; seems sensible idea. Regards, David Kernow 01:03, 24 April 2006 (UTC)
• Support, pretty straight forward. Shouldn't need a vote. —Pengo 07:01, 24 April 2006 (UTC)
  • Unforuntately the redirect has a (trivial) edit history so afaict it has to go through requested moves Plugwash 17:21, 24 April 2006 (UTC)

Done. —Nightstallion (?) ^{Seen this already?} 07:25, 28 April 2006 (UTC)

MTU versus TCP Window Confusion

The final paragraph of "Path MTU Discovery" has made a fundamental mistake regarding the implementation of Receive Window Tuning in Vista. The Receive window has nothing to do with the ethernet frame size or IP packet size (or MTU) but is in fact to do with the TCP window size - which is the number of TCP packets which may be transmitted before an acknowledgement is received. The paragraph which stands is correct - its just in the wrong place. —Preceding unsigned comment added by 87.194.2.137 (talk) 00:04, 28 January 2009 (UTC)

This is the paragraph that has been removed:

This problem has surfaced more frequently since the introduction of Windows Vista which introduces the 'Next Generation TCP/IP Stack'. This implements "Receive Window Auto-Tuning that continually determines the optimal receive window size by measuring the bandwidth-delay product and the application retrieve rate, and adjusts the maximum receive window size based on changing network conditions". This has been seen to fail in conjunction with older routers and firewalls that appeared to work with other operating systems. It is most often seen in ADSL routers and can often be rectified by a firmware update.

References

rename for better understanding

I would like to vote for remaning this article from MTU (networking) to Maximum transmission unit and to change the redirection the other way round.

I was just about to say the same thing. It seems pretty obvious that this page should be Maximum transmission unit rather than MTU. Richard W.M. Jones 21:08, 17 February 2006 (UTC)

Agreed.CecilWard (talk) 13:50, 30 January 2009 (UTC)

I like the style

It is true that this may not be the suitable Wikipedia style, but I like it and, as a student, I find it clear and useful. Doru001 (talk) 13:11, 20 January 2009 (UTC)

Does anyone have any comments about what is or was formerly wrong with the style/tone? I may be one of the offenders/improvers? Should the marker now be removed?CecilWard (talk) 13:49, 30 January 2009 (UTC)

Does anyone think that the descrition should refer to the fact that the MTU is the size of the PDU of the higher layer that is transported? This would make it clearer that the PDU of the transporting layer will be bigger than the MTU i.e. Ethernet II has an MTU of 1500 Bytes (of say IP) but the size of the Ethernet II frame will be larger than this at 1518 or more? Amore proprio (talk) 12:35, 6 May 2009 (UTC)

ATM section

I think the section about ATM optimisation sounds a bit wrong. For instance, the ethernet header (14 bytes) (I thought) would be stripped off before being sent via ATM. It also seems a bit irrelevant to be in Wikipedia - may be a personal website or something. --203.20.101.202 22:59, 26 November 2006 (UTC)

Ethernet header stripped off? Think again... how would you implement a layer 2 network over ATM then? —Preceding unsigned comment added by 81.241.160.113 (talk • contribs) 2007-07-29 21:05:40

My understanding is that with any WAN technology, the WAN would be the "layer 2 network". Ethernet headers are only used within the LAN (if Ethernet technology is used); layer 3 packets remain basically unchanged when going from one network to another, but layer 2 headers have to be stripped off every time the packet passes a router, and new headers - depending on the technology - added again for the next hop. Layer 2 does NOT offer connectivity between different networks (in the above example, Ethernet and ATM); that's what layer 3 (e.g., IP) is for.--Hilmarz (talk) 21:15, 31 December 2007 (UTC)

If the ATM network is using LANE, there will be an ethernet header. Otherwise there will not be, just raw L3 packets. I've seen LANE used, but only rarely. (76.118.216.33 (talk) 15:45, 11 September 2010 (UTC))

I agree with the earlier criticisms. I have expanded the section on ATM and added a concrete example of PPPoA optimal MTU choice which I hope is correct. I have marked _my own_ contribution as requires review by an expert.

I would also like someone to check and clarify the example of PPPoE (which PPPoE variant?) which I'm not sure about.CecilWard (talk) 13:06, 30 January 2009 (UTC)

ATM section is at very confusing or even wrong. I've attempted to rewrite it, but could not make it much better. While the "cell tax" is correct (although explanation is very bad - what is 'INT((payload_length+47)/48)' ??), the rest does not explain why changing MTU (which is MAXIMUM size) is a good thing at all. One can deduct that changing MTU for PPPoA connections is good because packet of MTU size are the most common (not true, large packets are only 20-40% of traffic) so it makes sense to optimize their transport (again not true, because for a 1500 byte packet 1 cell overhead is just 3%, while for a small 64-byte packet that make 40% of traffic there is 100% overhead). For PPPoE (_Ethernet_) 1492B MTU size (for PPP payload) is a result of maximum Ethernet MTU (1500B) minus 8 bytes space for PPP headers. It was not designed as optimized to fit ATM cells. What's more, MTU-sized packet will NOT fit exactly into 31 cells, because it is not 1492 or 1488 (detail: 1492 - 6 = 1486, not 1488) bytes! Headers are also transported (+6 bytes) and there is another 8 bytes for AAL5 PDU headers. That invalidates whole assumption because ATM payload has a much different size. —Preceding unsigned comment added by 64.103.25.233 (talk) 08:08, 4 November 2009 (UTC)

It think getting too far into detail in this article is a mistake. There is a huge variety of possible configuration scenarios for ATM which depend on what exactly is being carried on it (PPPoE versus raw IP packets versus whatever) and what encapsulation you are using (not everyone uses AAL5, AAL3/4 has also seen usage). Specific numbers are best left to network engineering literature. In addition this applies to cell/frame based networks in general and also to packet based networks where low MTU values on intermediate hops are present for other reasons (e.g. packet interleaving for QoS). ATM is, however, a convenient example to use. It would suffice to remove the section and just say "When large packets are fragmented, the majority of those fragments will necessarily be of MTU size. Cell based networks (or any underlying encapsulation layer which may cause additional re-fragmentation) incur less overhead if those fragments fit efficiently. As such, MTUs are sometimes "tuned" by network engineers to reduce overhead." (76.118.216.33 (talk) 15:45, 11 September 2010 (UTC))

MTU and Modems

For example, a 1500-byte packet, the largest allowed by Ethernet at the network layer (and hence over most of the Internet), ties up a 14.4k modem for about one second.

Shouldn't that be one tenth of a second? 134.76.88.99 (talk) 15:50, 3 March 2015 (UTC)

Nope. 1500 byte = 12,000 bit ≈ 14,250 serial bits (8N1). -- Zac67 (talk) 18:16, 3 March 2015 (UTC)

PDU vs SDU?

Article's first line says: "the size (in bytes) of the largest protocol data unit" -- but shouldn't it read "the size (in bytes) of the largest service data unit"? http://en.wikipedia.org/wiki/Service_data_unit — Preceding unsigned comment added by 174.61.224.227 (talk) 23:43, 12 January 2012 (UTC)

It is confusing, very confusing and so badly defined. But I explained this to myself this way: MTU of Ethernet is 1500, and 1500 is the max. payload which fits in Ethernet frame, then MTU is the largest PDU that comes from higher layer (IP) to the observed layer (Ethernet) that becomes its SDU and that it (Ethernet) can pass onward to lower layer. Hope I'm correct? Why didn't they just say "it's the maximum size of payload for the observed layer"?! --193.198.8.211 (talk) 17:54, 10 July 2014 (UTC)

PDU and SDU are only useful terms while you're talking about a single layer, otherwise they need to be layer specific (ie. "MAC PDU". MTU is a layer 3 term, so "PDU" is the correct term here – until you mention "Ethernet frame" which is L2 and the MTU becomes the SDU in L2... "Payload" isn't correct because the MTU is the maximum payload (SDU) of the N-1 layer. Zac67 (talk) 18:43, 10 July 2014 (UTC)

I don't agree with you Zac67. Something is fishy here. If MTU is maximal PDU of a protocol, than PDU of Ethernet is 1522 (or something). Maximal SDU of Ethernet is 1500. — Preceding unsigned comment added by 5.43.186.28 (talk) 00:11, 15 November 2015 (UTC)

The MTU is L3's maximum sized PDU. The L3 PDU becomes L2's (e.g. Ethernet MAC) SDU when it's transported. --Zac67 (talk) 11:15, 15 November 2015 (UTC)

Ethernet MTU

1500

Isn't it 1522 (2000 soon?) 87.207.141.240 (talk) 20:25, 24 June 2008 (UTC)

It depends how you define MTU. Internet RFCs tend to talk in terms of the MTU seen by the TCP/IP stack, that is the figure includes overhead from IP and higher layers but excludes overhead from lower layers. Plugwash (talk) 23:06, 24 July 2008 (UTC)

The length of the frame depends on what you count -- see the table/diagram at Ethernet#Physical_layer. 1500 is the max payload size for Type II Ethernet (1492 for 802.3 with SNAP.) Dunno about any proposed extensions.. -- 79.77.1.135 (talk) 23:11, 24 July 2008 (UTC)

What you count is defined in the RFC

Please have a look at the german MTU article which has good references to the rfc's defining the MTU. I'll fix it but right now i don't want to.

In computer networking, the term Maximum Transmission Unit (MTU) refers to the size (in bytes) of the largest packet or frame that a given layer of a communications protocol can pass onwards.

is wrong since or frame is not right, see german wikipedia article for references. —Preceding unsigned comment added by 134.2.186.16 (talk) 16:54, 12 February 2009 (UTC)

Agree with plugwash, indeed it is all about how your define MTU, which protocol layer and whether you are talking about the boundary immediately above or below that layer, (SDUs versus PDUs), and often confusion results from discussion including or excluding a header. Can we / do we need to improve the article in certain places in this respect?CecilWard (talk) 13:10, 30 January 2009 (UTC)

What about Ethernet "jumbo frames"? Any particular reason for not mentioning them? That would be 9000 bytes. — Preceding unsigned comment added by Dwandelt (talk • contribs) 18:36, 7 February 2014 (UTC)

The table tells that the MTU size is in bytes, while the reference is rather in octets, at least for the ethernet V2 line. Which one is correct? — Preceding unsigned comment added by 159.33.64.216 (talk) 19:44, 8 July 2015 (UTC)

These days, octets and bytes are the same thing. ~Kvng (talk) 14:12, 18 November 2015 (UTC)

Jabber

There is no explanation of why Jabber redirects here. Please add this. —Preceding unsigned comment added by 85.112.166.110 (talk) 08:49, 29 March 2011 (UTC)

 Done Jabber is now a disambig and Jabber (networking) redirects to Maximum transmission unit#Disruption where the term is in bold. ~Kvng (talk) 14:49, 22 December 2016 (UTC)

Jabber is a L1/L2 term, probably Ethernet only (check IEEE 802.3 8.2.1.5). So, in fact this isn't the right place for it as MTU is a pure IP/L3 term (IETF RFC 791). Should we move it to Ethernet frame? --Zac67 (talk) 16:00, 22 December 2016 (UTC)

The previous discussion on this talk page does not indicate a consensus for your assertion that MTU is exclusively a L3 concept. Have you found any secondary sources supporting your position? ~Kvng (talk) 15:01, 25 December 2016 (UTC)

MTU is defined in RFC 791 – what secondary sources are you requesting? Is there a higher authority on IP than IETF? Does any reference to MTU earlier or more authoritative than that RFC exist? Additionally, jabber detection is an MAU function defined for 10BASE5 (ie. it's a L1 function) – any better or earlier definition than that? --Zac67 (talk) 16:56, 25 December 2016 (UTC)

On Wikipedia, WP:SECONDARY sources are preferred over primary sources. They are considered more balanced and authoritative. I am looking for a source that says, "The IETF definition of MTU is golden and all other definitions are bogus." I hope you appreciate that we couldn't trust such a statement if it appeared in an RFC. This is one of the reasons why we prefer secondary sources. ~Kvng (talk) 20:09, 27 December 2016 (UTC)

For commercial sources, it's very reasonable to use secondary sources that are likely to be less exaggerating and more reliable. For a national or international standard however, there may be no (more) reliable secondary source. Since the standard defines the subject in question it should be considered the most reliable and authoritative source. Actually, this is common practice on pages on open network standards. Secondary sources are likely to be less reliable. Diluted terms like 'L2 MTU' and the like are the best examples imho. Probably I'm over-dramatizing the issue but I've had to find out the hard way that a vendor's documentation (secondary RS?) can be plainly wrong. Their mix-up of MTU and maximum frame size cost us a complete day of downtime. --Zac67 (talk) 22:45, 28 December 2016 (UTC)

The fact that a vendor's documentation was "wrong" is further evidence that there is not a strong agreement in the field as to the exact meaning of MTU. On Wikipedia, the way we handle this is to document (with sources) the different interpretations. It is not our role to try and correct the situation by declaring a single authoritative interpretation. Also, the IETF controls the definition of protocols, they do not control definition of TLAs. ~Kvng (talk) 15:27, 31 December 2016 (UTC)

Fair enough. Then we should reflect these different definitions in the text and not spread further ambiguity; as it is, the reader might very well be puzzled as to where the term applies and what that means. Any objection to section into something in the line of 'IETF definition (IP layer)' and 'Other definitions' as in Ethernet/L2? Jabber would belong there as well if not in Ethernet or Ethernet frame along with runt frames. --Zac67 (talk) 17:17, 31 December 2016 (UTC)

I went ahead and added MTU in other contexts here (to be expanded) and Jabber to the Ethernet page, as the latter is an L1/L2 error condition (at least originally) as can be seen by the included sources (it's just a timer, not a bit counter). If you're ok with it we should remove jabber here – exceeding the maximum allowed L3 PDU size is impossible since the underlying L2 won't accept it. --Zac67 (talk) 12:28, 4 January 2017 (UTC)

Thank you for that contribution (with citations!) ~Kvng (talk) 16:25, 8 January 2017 (UTC)

Glad you like it! You good with moving the jabber info from here to Ethernet as well? --Zac67 (talk) 17:21, 8 January 2017 (UTC)

Moved and I'm good with it. ~Kvng (talk) 15:28, 5 May 2020 (UTC)

Network Engineers Talking Among Themselves

I'll be the first to say it: Lots of people come here because there's an MTU Size adjustment for their routers, and they don't know the first thing about what do with it, except that if adjusting it might improve their throughput, they'd like to maximize it.

But, before the first paragraph in the article is finished, the direction of the content diverts from Basic Definition of MTU to assorted technical details, such as how MTU varies with different systems, etc.

I would like to see more dwelling on the Basic Definition for those of us coming here with our router-concerns. For instance, MTU Size is on the Basic Settings page of my router's web-interface, so I come here expecting to be able to find a description that will round out my Basic Understanding. That would require mixing in some Customer-Service perspective into all of this Engineering talk, which obviously requires talents that go beyond engineering.

I believe my comment correlates with why this article has a "WikiProject Article Quality Grade of "C."

Thanks, Nei1 (talk) 15:47, 29 June 2011 (UTC)

The current second paragraph of the lead does a good job with this now. The version at the time of this comment was unnecessarily complex. ~Kvng (talk) 15:28, 5 May 2020 (UTC)

802.11 MTU

I'm finding conflicting and confusing information for 802.11 MTU. Anyone have have a definitive reference and explanation? --Kvng (talk) 15:59, 23 December 2011 (UTC)

See the 802.11 2012 Specification on page 413, section 8.3.2.1 - Data Frame Format — Preceding unsigned comment added by Roy muzz (talk • contribs) 01:42, 25 May 2012 (UTC)

At the time I'm posting this, the article claims that the 802.11 MTU 7981 bytes. However, all the 802.11 specifications show the frame size to be 2346 bytes (34 byte header and 2312 byte body). Furthermore, the reference points to page 413, which is a flowchart, and has no MTU information at all. — Preceding unsigned comment added by 209.6.233.169 (talk) 20:57, 29 January 2016 (UTC)

Same user as above; I take back some of what I said. The 802.11 reference is correct, for the 2012 publication of the same. The older publications with the same name have different content. — Preceding unsigned comment added by 209.6.233.169 (talk) 21:00, 29 January 2016 (UTC)

I have added the 802.11-2012 ref to the 2346 figure in the table. ~Kvng (talk) 15:28, 5 May 2020 (UTC)