Realtime emulation scalability:

[S.V.R. Anand]

Hello,

I am building an ns-3 emulation testbed using RealtimeSimulatorImpl and EmuFdNetDevice to mirror a physical multi-BSS WLAN deployment, and I am trying to understand the scalability limits of the setup.

Configuration
ns-3.45
optimized build (--disable-examples --disable-tests)
802.11ac, 20 MHz orthogonal channels: 36 / 40/ 44 / 48
one YansWifiChannel per BSS. MinstrelHtWifiManager
2 spatial streams
APs bridged through a BridgeNetDevice onto a CSMA backbone
one STA per client acts as an IP router with an EmuFdNetDevice connected to a veth pair whose peer resides in a Linux network namespace
server also runs in a namespace
no traffic traverses a physical NIC

Host machine

Intel Core i9-14900

Max frequency: 5.8 GHz

ns-3 running single-threaded

iperf3 --reverse. DL TCP.

`

What I have seen is `

2 AP × 4 STA

Worst wall-clock lag:      3 ms

Mean MPDUs per A-MPDU:    31.9

Aggregate throughput:    227.8 Mb/s

4 AP × 2 STA

Worst wall-clock lag:     65 ms

Mean MPDUs per A-MPDU:     2.3

Aggregate throughput:    149.4 Mb/s

I also find that ns-3 is lagging behind the wall-clock time  in the 4AP - 2STA case.

Can I do something to make the ns-3 emulator support more APs and STAs ? 

Regards

Anand

[Gabriel Ferreira]

You can do some profiling to see what is slowing you down.

At these rates, I'd guess your problem will be fixed with https://gitlab.com/nsnam/ns-3-dev/-/merge_requests/2969

[Tom Henderson]

A few other suggestions:

- make sure that offloads are disabled (run 'ethtool -K <veth> gro off gso off tso off' on every virtual device).  If offloads are on, the kernel could be delivering jumbo frames to the packet socket which may drop them.  This would contribute to lower throughput.

- if your BSS are independent from an RF perspective (you are not trying to model interference across BSS), you could consider to put each AP and its associated STAs in a separate ns-3 process, and replace the CSMA that interconnects them with a FdNetDevice on a Linux bridge.

- the aggregation in 4 AP x 2 STA case (note, I don't exactly know what your notation means here) drops significantly.  This probably means that the Wi-Fi queues are shallow, perhaps due to TCP getting bursty and operating with small windows, and it may be a feedback effect (as the simulator gets late, the TCP RTT increases, leading to larger BW-delay product, and lower TCP throughput...).  This is just a guess at why the aggregation is low.  Perhaps try also to turn on A-MSDU (BE_MaxAmsduSize) and see if you have improvement.

- Tom

--
Posting to this group should follow these guidelines https://www.nsnam.org/wiki/Ns-3-users-guidelines-for-posting
---
You received this message because you are subscribed to the Google Groups "ns-3-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ns-3-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/ns-3-users/212de1d7-1b7a-4dfb-893d-03b8ec4d0aecn%40googlegroups.com.

[Anjali Shaw]

I read multiple post and consulted multiple resources but the only way i could find to reproduce a network in ns3 is by doing manual calibration, is there any other way to reproduce the same ? One way I could think of is by path loss models but I assume that it would concern other modules as well. Kindly suggest what can I do 

For a context, I am trying to simulate the env to do resource allocation on application layer so is there any other tool as well that could help me better ? 

Can anyone help me with this ?