[LENA] Using TCP sees more link layer ReTx than UDP

[Xing Xu]

Dear all,

For the same setting (described below), if I use TCP for one UE (target UD), I saw many more link layer ReTx, than using (bi-directional) UDP for that UE (in AM mode, of course). For both cases, the BS is pretty busy for DL, which makes the comparison valid, see below.

Configuration

My configuration have 3 BSs work on the same frequency, the UE I focused on belongs to 1 BS, the other 2 BSs are interference (this 3 BS forms one cell, using LteHexGridEnbTopologyHelper). Interference have both UL and DL. All 3 BSs are congested for DL. My target UE either uses TCP, or uses bi-directional UDP. For both cases, the DL traffic can congest the BS; the UL traffic for UDP setting is not heavy, the purpose of adding it is to simulate TCP's ACKs traffic on UL, to compare to the TCP experiment.

Result

I log ReTx at lte-rlc-am.cc. I observed a lot ReTx happens for the TCP run (hundreds of them in the single run) but not the bi-directional UDP one (<5 in the single run).

I don't know the reason of that... As I think for both cases, the interference is the same, DL traffic for TCP and UDP should be the same as they both congest/overload the BS; for UL, I'm not sure about the workload, but the difference would not be so big: I mean, if we'll see lots of UL interference and thus lots of UL ReTx, I should observe similar amount of ReTx for both TCP and UDP case. Actually, if I add more traffic on UDP setting's UL, I didn't see more ReTx for UDP setting, so I guess in short, I see TCP setting has lots of link layer ReTx however for (bi-directional) UDP there is no such ReTx.

I don't know how to debug from this point. Can you tell me how do I differentiate UL ReTx and DL ReTx? Currently I'm logging at lte-rlc-am, so I guess I logged the aggregated ReTx for UL and DL. Also, how can I know the root cause of such ReTx (packet loss)? E.g., can I get to know how did each packet loss happens? Again, for TCP and UDP packets, at link layer, they are just data, I don't see how they differ at link layer and why there are so many packet losses for one setting but not the other.

Thanks,

Xing