3

With the various DSL flavours such as ADSL2+, which allows the PHY line to re-sync as conditions worsen and improve at higher and lower sync rates (which is effectively Tx and Rx rate), does the MTU size ever change? I know that it is using fixed size ATM frames here, but will the MTU size ever drop below the minimum ATM frame size?

Also, what about Wireless (802.11)? As the wireless signal becomes weaker and the error rate increases, is the MTU ever shrunk so smaller frames are sent?

I understand that at an the IP level the window size can be shrunk/grown as with the MSS at the TCP level for example. But what about at OSI layer 2, using Ethernet and ATM for example, are there any scenarios where the MTU size would ever change over the course of a link's life? I don't mean over 20 years either, I mean within a reasonable time span, such as days or weeks?

  • 2
    Hi, could you provide some background information around the reason for the question... are you solving a problem? If so, what is the problem? Why are you talking about both ADSL and wifi MTU in the same question? Finally, there are a few different ways to slice MTU information... are you talking the size of an IP packet, or something else? – Mike Pennington Aug 12 '13 at 21:21
  • @MikePennington The reason for the question is not relevant, the question is relevant within the site's guidelines. I mention both ADSL and Wifi MTU because these are layer 1 technologies atop of which sit layer 2 protocols, which independent of the layer 1 protocol have an MTU size. I am not talking about the size of IP packets, I mentioned already that Window size and MSS change, and that I'm referring strictly to layer 2 MTU. Not sure why you posted the above questions, seems like unnecessary/pedantic criticism to me, even if it has the best intentions. All answers are in my original post. – jwbensley Aug 13 '13 at 9:38
  • 4
    I'm asking for clarification to help you... this question pastes together so many disconnected concepts a reasonable person asks himself if there was a coherent bigger picture. I apologize if you hear criticism in the question, I'm simply trying to a) help you express the need and b) perhaps help myself or others find a solution to the bigger picture in your question. It is very common for questions that come here to have incomplete information. – Mike Pennington Aug 13 '13 at 10:45
  • Did any answer help you? if so, you should accept the answer so that the question doesn't keep popping up forever, looking for an answer. Alternatively, you could provide and accept your own answer. – Ron Maupin Aug 8 '17 at 14:22
4

An ATM cell is 53 bytes. Using AAL5, there's 48 bytes of payload. With the minimum ethernet frame of 64 bytes, you are unlikely to ever find an MTU smaller than an ATM cell. Even if you did, every protocol I know of with a fixed minimum includes padding to get to the minimum.

MTU is an interface setting. It is not negotiated. It does not change over time. Your MTU will be what you set it to (or some default (1500, 4470, etc.)) until you change it.

| improve this answer | |
  • I'm not entirely sure what relation ATM and ETH have here, but I'm assuming people are trying run ATM over ETH. ETH has minimum payload of 46B, so you couldn't fit ATM cell inside minimum sized ETH frame. – ytti Aug 13 '13 at 7:24
  • @Richy Beam Ah, I overlooked negotiation of a changing MTU. I was asking the question because I wondered if any protocol might lower it's MTU if for example the quality of the physical link layer degraded significantly (that is why I mentioned ADSL, because it is auto adjusting). However, if one end of the link where to lower it's MTU, the other end must also conform in the same way. That is something I had overlooked. After more research and the answers here I can see that MTU is fixed for all protocols, and having a changing value would be problematic. – jwbensley Aug 13 '13 at 9:47
  • @ytti, 99% of IP traffic is going to originate from an ethernet host. The minimum frame is 64bytes (mac thru fcs), or 46byte payload (sans vlan.) In fact, linux won't allow an mtu less than 68. How ETH-ATM interact depends on the protocol. It's a moot point as almost everything supports padding to meet any minimums. – Ricky Aug 13 '13 at 23:32
  • Linux MTU is L3 MTU (Like Cisco IOS), standards won't allow it lower than 576B for IPV4 or 1280B for IPv6. Fully agreed it would be hard to find practical system (as all target IP) where ATM won't fit in ETH, but ETH (L2 MTU) happily could be less than ATM cell. – ytti Aug 14 '13 at 6:10
3

Also, what about Wireless (802.11)? As the wireless signal becomes weaker and the error rate increases, is the MTU ever shrunk so smaller frames are sent?

The 802.11 specification does not suggest to shrink the MTU in case of bad links. But the layer 2 isn't independent of the frame size: The option to use RTS/CTS or layer2 fragmentation/aggregation depends on the frame size. The chosen modulation rates could also depends on the frame size.

And since the layer 1 is a shared radio environment, it's packet loss probability may depend on the size of the frame. You could theoretically have a lossy environment where larger frame have a higher loss probability than short frames, but you could also have the opposite. This depends on the radio environment, the amount and type of noise or competing radio technologies in use. But if you use Wi-Fi in ESS/infrastructure mode, the association will be dropped well before you get to that point.

| improve this answer | |
2

The only dynamic adjustment of MTU size is to avoid IP fragmentation. That is changing the TCP segment size, to match the smallest IP packet size. IP packets can be divided up by say multilink PPP but that only lasts for the single MLPPP hop.

Bottom line no protocol ever changed sizes to reduce retransmissions it was said in the old days that you might want to manually adjust data sizes down when using bad links but we are talking at least 15 years ago.

| improve this answer | |
  • Thanks for the info. Be careful with your wording The only dynamic adjustment of MTU size is to avoid IP fragmentation - MTU size is used to refer to layer 2 frame maximum transmission unit size. Layer 2 MTU isn't changed to avoid IP fragmentation. As you go on to mention TCP, I believe you mean MSS. – jwbensley Aug 13 '13 at 9:43
  • True I meant MSS (max segment size) – fredpbaker Aug 13 '13 at 11:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.