1

I don't mean calculating minimum MTU along a path.

I mean, what factors determine how large an MTU can be at all? Is there a physical bottleneck of some sort? For example MTU for the Internet is 1500 bytes. How did we arrive at this number? I know it's the size of an Ethernet Frame, but why is that the maximum size? What happens if we try to send more than 1500 in one frame, which was decided to set as the frame size?

Improve this question
4
  • There is no fixed internet MTU Commented Aug 15 at 17:02
  • 1
    The ethernet payload is 1500, hence the MTU for an ethernet connection is 1500. Wi-Fi has a larger payload, as do most serial type links and token ring. The minimum in the path is your limitation. Commented Aug 15 at 17:08
  • @RonMaupin yes, I understand this fact that the minimum in the path is the limitation, my question is why does the the minimum in this case the ethernet I guess have a limitation of 1500 bytes. A larger frame size would mean more efficient communication, so what drawback occurs for frames > 1500 bytes to cap the frame size at 1500. Commented Aug 15 at 18:10
  • The ethernet standard, as defined by the IEEE is a maximum payload size of 1500. Other layer-2 protocols have other maximum payload sizes. Sending a frame larger than the standard will simply get it dropped when it reaches a standard ethernet interface. The receiving standard interface will have it reported in the hardware counters as damaged and dropped. Commented Aug 15 at 20:52

3 Answers 3

Reset to default
2

You didn't specifically ask about Ethernet, but seeing how almost everything has become ethernet today, I'll assume that's what you're interested in. The historical trivia of "why 1500" has been asked and answered several times... they needed a number and that's what they picked. Though there are/were very logical reasons, it was an arbitrary number.

If you're asking about Jumbo Frame - anything larger than the 802.3 specified 1500, then there is no standard - 802.3 refuses to define/support it. Thus the only meaningful limiting factor is whatever ALL of your hardware can support. Increased latency and jitter due to transmission delay is why we generally shy away from larger frames on slower links (like 10 and 100), but with modern speeds (100...800+), very large frames don't make as much difference.

Improve this answer
3
  • The historical trivia of "why 1500" has been asked and answered several times... they needed a number and that's what they picked. Though there are/were very logical reasons, it was an arbitrary number. I'm sorry this is exactly what I'm looking to understand, if possible can you give me some resources that talk about the reasons (apart from the arbitrary choice). I'm learning about computer networks and wanted to understand why the number was chosen. Commented Aug 21 at 16:16
  • I couldn't say. This long pre-dates the internet and google. If it is written down, it's likely hand written notes from an ancient 802.3 meeting. Or even further back before 802.3. Remember this was in the days of 10b2 and 10b5, so there were hardware considerations as well. (eg. 802.3as notes about extending ethernet to 2k frames) Commented Aug 22 at 5:08
  • I don't know the exact reasoning, but bear in mind that Ethernet as we know it appeared around 1980. 1500 bytes seems tiny in modern terms but not so tiny in 1980 terms. Commented Sep 26 at 15:37
2

Technically, the maximum frame size for a specific link layer protocol is set by the standard, as a tradeoff between cost (for buffers and processing capability) and efficiency (larger packets mean less overhead for processing and transport). Once set, the standard usually doesn't change, even after decades.

That said, it's effectively you who decides which MTU your network can carry. You decide by choosing the components that connect your network and which protocols they speak.

If you choose the almost ubiquitous Ethernet and use standard frames, then that MTU is 1500 bytes. If you use non-standard Ethernet, the MTU could be 9000 bytes or even higher. If you run Wi-Fi throughout, then you'd get 2302 bytes.

For your Internet access, you could possibly negotiate a larger MTU than 1500 bytes with your ISP, depending on what they offer (between sites using that same ISP). Without negotiating or special offers, 1500 or 1492 (with PPPoE or PPTP) is probably what you get.

Beyond your ISP, 1500 bytes is extremely likely all that's possible, due to nearly all peering being done with Ethernet.

Improve this answer
1

You won't find some fundamental physical reason, just a sad story of engineering tradeoffs, and ossified standards.

In general, smaller packets have a couple of advantages.

  • Latency/Jitter. In most cases a packet must be received before it is forwarded (cut-through switching does exist but has it's own issues). So with larger packets each hop adds more delay. Furthermore until a packet is completely transmitted nothing else can be sent out, even if it's higher-priority.
  • Less buffer memory needed, any device in the network needs to be able to buffer at least one packet and preferably multiple. Not a huge deal nowadays for full-sized computers, but a much bigger deal historically and still a big deal for embedded applications.

On the other hand larger packets also have a couple of advantages.

  • Less encapsulating overhead, leading to more usable bandwidth out of a link.
  • Less forwarding descisions need to be made, this is a major issue for high-speed networks where a switch or router may need to make hundreds of millions of forwarding descisions per second.

In general, faster networks and more powerful computers favour larger packets, while slower networks and smaller computers favour smaller packets.

Which raises the question, why has the MTU for general-purpose Ethernet networks remained the same for nearly 50 years, while networks have got so-much faster? and why has that become the de-facto "internet MTU"? Well there are a few factors.

  1. "IP over Ethernet" has no mechanism for detecting or negotiating MTUs. So running an Ethernet network with a larger than default MTU requires manual configuration. If this manual configuration is not done consistently or if there are legacy devices that can't handle the larger packets then packets may simply dissapear.
  2. While IP has mechanisms for handling mixed MTUs along a forwardin path none of them are problem-free. Fragmentation causes overhead in extra forwarding descisions and in processing to reassemble the fragmented packets. Path MTU discovery causes extra latency as the ICMP messages return to the sender, and can also break if ICMP messages are filtered.
  3. The benefits from a larger MTU only come when the entire path supports the larger MTU. As an ISP if your customers lans only support 1500 bytes, and the service providers that feed you with content only support 1500 bytes, what incentive do you have to support more than 1500 bytes?
  4. While it hasn't necessarily been cheap or easy, makers of routers and switches have managed to keep up with the increasing data rates and produce devices that can forward typical internet packets* without getting backlogged.

* The average packet size will be smaller than the maximum due to acknowlegement packets, connection setup packets, packets only carrying a small amount of data and so-on.

Improve this answer

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.