What was the protocol usually used on top of X.25 and ATM or even Frame Relay Networks when those were commonly used?

Was it also TCP/IP protocol stack or some other protocol stack?

From what I know since the beginning of the Internet (ARPANET, NSFNET) TCP/IP protocol stack was widely used at least on the access layer of the networks (host machines).

However I know that X.25 and ATM networks were also widely used but I am not sure at what level did those networks work. Were they also carrying TCP/IP traffic?

  • Unfortunately, questions about historical trivia that "is trivial/irrelevant to modern networking" are off-topic here.
    – Ron Maupin
    Sep 24, 2017 at 16:27
  • 3
    This Q is here since 5 Sept. I take an evening of my time to answer it, and minutes later the question is put on hold. This is deeply demotivating when it took an hour of my evening to write a reasonable answer. I don't disagree with the closure, but you'll appreciate that these circumstances don't encourage contributions. Best wishes, glen.
    – vk5tu
    Sep 24, 2017 at 16:46
  • @vk5tu Hey thank you for answering. I appreciate the answer. I hope you do not get demotivated by the fact that my question was put on hold. I think the moderation on this website is wayyyyy to exaggerated .
    – yoyo_fun
    Sep 28, 2017 at 8:00
  • @vk5tu Putting it on hold doesn't prevent me from upvoting your answer. :)
    – BobRodes
    May 17, 2020 at 8:17
  • The Internet has always been IP. Note that it's called Internet Protocol. An Internet without Internet Protocol is not the Internet, just a big network.
    – user253751
    Feb 15, 2021 at 15:25

1 Answer 1


Think of it this way: we used the fastest links we could buy to run IP over the wide area between the nodes of the Internet. So that was 56/48Kbps X.25, and later 1.44/2Mbps Frame Relay, and later 155Mbps ATM.

The owners of those links used them for other things too. In fact they usually despaired of their use for IP. They saw that as "commoditising the long-lines business", as IP could run over any transmission technology without its users noticing, and so there was no lock-in to any particular transmission technology or vendor.

X.25 had a set of related protocols. A user wanting to connect to a remote site would sit in front of a VT100-style terminal, which was wired to a X.29 "Packet assembler/disassembler" (PAD). That would produce a X.25 connection to the remote site, which would be established via a call through a set of X.25 switches. After finishing with the remote computer the call would be cleared down and the user would be billed for the minutes and bytes used. IP traffic did not fit well into the connection-oriented business model which X.25 assumed, and so IP over X.25 was always an uncomfortable fit.

Frame Relay and ATM were much more transmission-oriented technologies. They didn't define application-style services, that was left to another set of protocols called "OSI". Rather ATM had a set of "adaption layers" for each type of traffic it supported: packets, T1/E1 trunks, Voice calls, television signal distribution, and so on. The ability to run all these types of traffic on a single long-line link was seen as a great advance by the telco's sales people, as they could sell any service a customer might order without needing to wait for a transmission to be built which was particular to that service.

But remember the IP people didn't care about such things. They simply bought the technology with the highest economic throughput, as their bandwidth needs were expanding by 40% per year. The moment there was access to SONET/SDH then ATM was dropped as overly complex and thus too expensive for 1.2Gbps compared with 1.2Gbps SONET/SDH. The same fate happened to SONET/SDH when Gigabit Ethernet used the very simple Fibre Channel optical encoding.

These days long-haul IP traffic runs in the DWDM channels of the optical fibre technologies used for long-line networking. To reduce costs there is interest in router interfaces producing the frequency of light and forward-error correction to match that used by the DWDM system, then the router's signal can by optically mixed into the long-haul signal with no further processing.

For long-haul IP we're not even buying Gbps anymore, we're buying a spectrum of light within a optical analogue transmission long-line system. How many Gbps that turns out to be depends how wide and complex a signal you place within the bounds of that spectrum allocation, and what error rate is acceptable to you.

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