What could be issue of low TCP throughput in DL 5G SA?

Hi Experts.

What could be issue of low TCP throughput in DL 5G SA?

I see BLER is fine and everything is OK.

MCS 26-27, 273 RBs is always used.

But somehow I see maximum TCP throughput is 700 Mbps, which it shall be 900 Mbps.

What could be the reason?

UDP is fine and reaching 950 Mbps, delay of ping also on the system is fine.

What’s the MIMO status? 3 or 4 layers?

4 layers, RI 4.

That is RI of the UE, but are you sure you have 4 PDSCH layers at gNodeB?

Yes, of course.

What happen is that at first timestamps of tcp running I see like 900 Mbps, then quickly it drops to 700 Mbps.

It seems to me you are in a Lab environment.

There can be many reasons.

Do you have enough data coming from higher layers in DL?

If no, check that enough data is coming form Core.

If yes check BBU limitations (CU) or RU (DU) limitations.

Sometime in Transmission there is flow control that limits throughput.

Used -P 15 parallel streams.

What are HARQ retransmission rate when TCP vs UDP? (HARQ 1/2/3… rate)

Is it a real UE or simulator ?

You can check the congestion control algorithm (Cubic or Reno) and TCP window size, buffer size.

It’s bbr. Already checked on Core side is bbr.

Real UE.

-u -b 100M -P 10 -R meaning you ask for 1 Gbps in Downlink.

Then compare tcp dump from s1 -u and UE side tcp flow, there could be a problem either at Pdcp not enough data coming losses on S1 or at Ue side app layer.

Do I need to specify the -b when tcp used?

I think -b on iperf on udp no?

Yes, I said for UDP.

At cell on its pdcp I see max arrive 700 Mbps.

But remove -b and make tcp parallel streams.

What is the bw of cell?

100 Mhz, 273 RBs used, MCS 26.

BLER is fine not high.

Ok, so you should see 1.6 - 1.7 Gbps.

But in rank4 condition.

If prbs max utilized and mcs is high with rank4, in tcp there is slow start mechanism so your mcs increases, than if there is packet losses gnb degrades the mcs assigned.

Most probably ur case is related with Transmission.

I recently testing FR2 800 Mhz.

I replaced tx from 3rd party backhaul to dark finer.

Fiber, throughput hit the 3.5 Gbps.

Actually it’s like this: I have 2 5G SA cells, both same are identical on all configs .

  • Cell1 connected to Core 1
  • Cell2 connected to Core 2

Cell1 - udp in DL is 950 Mbps, DL tcp around 900 Mbps. (It’s working fine)

Cell2- udp in DL is 950 Mbps, DL tcp around 700 Mbps.

Both cells has mcs 26, 273 RBs, everything is same in terms of RAN configs.

On cell2 (problematic cell) I see that initially there is on tcp like 950 Mbps, then after little timestamps it downgraded to 700 Mbps in tcp DL.

I see than Ng user plan sdap losses to Upf can be considered.

This degradation mostly because of packet loss on tcp.

What do you mean here?

Loss comes where?

On cell itself what arrives it transmit …

So no incoming packet coming more than 700 Mbps from Core.