Last Post: September 3, 2005:
- If you are in a noisy RF environment, one of the possible solutions is to implement fragmentation. This is done because of too many retransmissions. But, won't TCP start fragmenting with sliding window if it isn't recieving the proper TCP ACKs? I have some theories about this, but nothing solid.
- Ooo... another comment/question. If a wired node is fragmenting (for whatever reason) will the AP reassemble prior to sending to the wireless STA? And, if the wireless STA is fragmenting at the MAC layer, does the AP reassemble prior to sending out to wired? I'm pretty sure that the answer to both is yes, but wanted to confirm.
- However, if the MAC layer is having a lot of retransmissions, doesn't it become pointless at some time (lets say after the 3rd try) because the TCP layer didn't get it's ACK in time where then it will retransmit anyway?
If anyone has comments, answers or references, I would much appreciate it.
IEEE 802.3 Ethernet has no fragmentation and reassembly mechanism.
IEEE 802.11 fragmentation and reassembly takes place between a transmitter and its immediate receiver. The immediate receiver finishes reassembly before taking any forwarding action.
IP has a fragmentation and reassembly mechanism, and has no acknowledgment or retransmission mechanism. The originating host and every router along the path may fragment packets into smaller more numerous packets. Only the destination host reassembles.
TCP has no fragmentation and reassembly mechanism, and has an acknowledgment and retransmission mechanism. The retransmission timeout is notoriously slow.
IEEE 802.11 retransmission after a failed ACK, whether of a fragment or an entire frame, is much faster than TCP's retransmission timeout. Even multiple 802.11 retransmissions of the same frame or fragment, so long as the frame is not discarded for too many attempts, will not trigger a TCP retransmission.
I hope this helps. Thanks. /criss
Thanks for the reply. I have to study up on layer 3 and 4. I better go find my O'Reily TCP/IP book....