4

Why is flow control implemented in both transport and data link layer? If it is already implemented in the data link layer then what is the need to implement it again in the transport layer.

1
  • 1
    Not long enough for an answer, but one is the macro (entire TCP conversation and loss/endpoint buffers) and one is the micro (physical network link and per-interface buffers).
    – cpt_fink
    Jun 26 '15 at 6:05
6

Why is flow control implemented in both transport and data link layer?

Because flow control is not implemented in all transport layer protocols.

TCP may be popular, but there are a non-trivial number of people using UDP applications, or completely non-IP applications (such as FCOE, which heavily depends on flow control in DCB / Lossless ethernet)Note 1.

Add to this cpt_fink's comment that data-link flow-control can be implemented as per-hop behavior (which basically goes back to my first point). By definition, transport flow control cannot be implemented thusly.


Note 1 I'm personally not fond of the notion of "lossless ethernet", but I can't change the reality that people use that term.

2
  • Please explain this :"data-link flow-control can be implemented as per-hop behavior ".
    – Alex_ban
    Jun 26 '15 at 14:23
  • I am blatantly abusing the PHB term, since it officially applies to diffserv / mpls. It was an lazy way of pointing out that flow control can be implemented independently at each switch hop
    – This
    Jun 27 '15 at 8:48
4

Actually, you must take into account that the TCP/IP protocol was not built to run over ethernet exclusively. There are various link-layers that might be used with TCP/IP. Some might provide a local link-only flow control (like between two serial ports using XON/XOFF) or something like that. Thats why the TCP protocol ended up providing a flow control mechanism that works at a much higher level and is generic enough to traverse multiple different types of networking links.

2

Flow control at the data link layer extends only as far as the layer2 link. There are typically several layer 2 hops between IP endpoints.

Flow control at the transport layer (TCP) extends end-to-end and accommodates communication between NICs/endpoints with different throughput/speed.

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.