Topic packets from Windows to WSL interface dissappearing (ROS2 app)

[borjamunozf]

Windows Version

10.0.22631.4037

WSL Version

2.2.24

Are you using WSL 1 or WSL 2?

• WSL 2
• WSL 1

Kernel Version

5.15.146

Distro Version

Ubuntu 22.04

Other Software

WSL version: 2.2.4.0
Kernel version: 5.15.153.1-2
WSLg version: 1.0.61
MSRDC version: 1.2.5326
Direct3D version: 1.611.1-81528511
DXCore version: 10.0.26091.1-240325-1447.ge-release
Windows version: 10.0.22631.4037

Repro Steps

• Using CYCLONE_DDS = - rmw_cyclonedds_cpp
• sudo apt install ros-humble-ros-base

Expected Behavior

Packets are not lost in WSL & performance are the expected (~16Hz)

Actual Behavior

Three developers in my company has encountered strange performance issues in an application. This does not ocurred before without WSL!

The scenario is following:

• Windows Unity application.
• WSL 22.04 Ubuntu running ROS2
• Wireshark analyzing WSL hyper-v adapter from
  From Windows side all looks good, the topic is published every ~6ms like a clock.
  

But from WSL a lot of packages are missing. It's only receiving like 1 update per second on average.

I have established the /proc/sys/net/core/rmem_max" to higher value = 2203936 and added this to the kernel:
CONFIG_IP_MULTICAST=y

It seems that this has slighly improved performance. However It starts really nice and afterwards it just get bad again.

Diagnostic Logs

No response

[github-actions]

Logs are required for review from WSL team

If this a feature request, please reply with '/feature'. If this is a question, reply with '/question'.
Otherwise please attach logs by following the instructions below, your issue will not be reviewed unless they are added. These logs will help us understand what is going on in your machine.

How to collect WSL logs

Download and execute collect-wsl-logs.ps1 in an administrative powershell prompt:

Invoke-WebRequest -UseBasicParsing "https://raw.githubusercontent.com/microsoft/WSL/master/diagnostics/collect-wsl-logs.ps1" -OutFile collect-wsl-logs.ps1
Set-ExecutionPolicy Bypass -Scope Process -Force
.\collect-wsl-logs.ps1

The script will output the path of the log file once done.

If this is a networking issue, please use collect-networking-logs.ps1, following the instructions here

Once completed please upload the output files to this Github issue.

Click here for more info on logging
If you choose to email these logs instead of attaching to the bug, please send them to [email protected] with the number of the github issue in the subject, and in the message a link to your comment in the github issue and reply with '/emailed-logs'.

View similar issues

Please view the issues below to see if they solve your problem, and if the issue describes your problem please consider closing this one and thumbs upping the other issue to help us prioritize it!

Open similar issues:

• Network packet discrepancy/drop between Windows Host and WSL2 (#10989), similarity score: 0.74

Closed similar issues:

• IP protocol traffic not registering on WSL2 network interface (#6831), similarity score: 0.74
• DNS packets seemingly randomly dropped (#10258), similarity score: 0.72
• Packet loss between WSL2 container and host (#5530), similarity score: 0.71
• WSL2: Do not recieve UDP packets from network interface using mirrored option (#11027), similarity score: 0.70

Note: You can give me feedback by thumbs upping or thumbs downing this comment.

[borjamunozf]

Interesting update:

• Setting up the same scenario in two laptop with Windows 10, WSL 2.2.4/2.1.5 and the standard kernel (or the customized one, 5.15.146.1) we get roughly expected performance and packets.

The laptops with expected performance:
Laptop 1

• Windows 10.0.19045.4780 - WSL 2.1.5
• 12th Gen Intel(R) Core(TM) i9-12950HX
  Logical CPU cores: 24
  Has AVX512: no
  L1 cache size: 48 KiB
  L2 cache size: 1280 KiB
  L3 cache size: 30 MiB
  L1 cache sharing: 2 threads
  L2 cache sharing: 2 threads
  L3 cache sharing: 24 threads
• NVIDIA RTX A3000 12GB Laptop GPU

Laptop 2

• Windows 10 Enterprise 22H2 19045.3570 - WSL 2.2.4
• Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz
  Logical CPU cores: 12
  Has AVX512: no
  L1 cache size: 32 KiB
  L2 cache size: 256 KiB
  L3 cache size: 12 MiB
  L1 cache sharing: 2 threads
  L2 cache sharing: 2 threads
  L3 cache sharing: 12 threads
• NVIDIA RTX A3000 12GB Laptop GPU

Right now we're considering issues with WSL, Windows 11 and Intel P-E cores as the faulty laptops uses 13th Gen Intel(R) Core(TM) i9-13950HX

Not sure if could be connected to #9190

[CatalinFetoiu]

thanks for reporting the issue. can you please collecting networking logs using instructions at https://github.com/microsoft/WSL/blob/master/CONTRIBUTING.md#collect-wsl-logs-for-networking-issues

the script will generate a zip with name starting with "WslNetworkingLogs"

[borjamunozf]

thanks for reporting the issue. can you please collecting networking logs using instructions at https://github.com/microsoft/WSL/blob/master/CONTRIBUTING.md#collect-wsl-logs-for-networking-issues

the script will generate a zip with name starting with "WslNetworkingLogs"

The problem is #11961 In my company all laptops has a specific and different admin user account, so opening a Powershell and running the scripts as admin does not detect the WSL distributions because they're installed in the normal account.

[CatalinFetoiu]

@borjamunozf thanks for following up. is there a way to install and reproduce the problem on the admin account?

[borjamunozf]

I'm modifying the script to adapt to our environment. Hopefully I'll be able to run the script with some changes and upload the logs on Monday! El sáb, 21 sept 2024 a las 2:20, Cătălin Emil Fetoiu (< ***@***.***>) escribió:

…

@borjamunozf <https://github.com/borjamunozf> thanks for following up. is there a way to install and reproduce the problem on the admin account? — Reply to this email directly, view it on GitHub <#11966 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AGFV5D5TGQYLDYEKLBAGNCDZXS3VRAVCNFSM6AAAAABNINS4LOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNRUG43TMMZVHA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[borjamunozf]

Ok, this is the logs we have been able to collect in one laptop. It has been quite tricky to make it work with our company environment and account policy :/ probably some data has been not possible to gather.

• The laptop in this case is a Windows 11 with 2.2.4 WSL
  13th Gen Intel(R) Core(TM) i9-13950HX

Started application to collect data with the tcpdump, extremely poor performance (not even ~1Hz this time...)

WslNetworkingLogs-2024-09-24_13-58-42.zip

We'll try to get the same logs but in Windows 10 with 2.2.4 WSL and 12th Gen Intel(R) Core(TM) i9-12950HX
that seems to be working well enough.

EDIT: It seems that something it's missing... Will update the logs again as soon as I fix this.

[github-actions]

No logs.etl found in the archive. Make sure that you ran collect-wsl-logs.ps1 as administrator and that the logs.etl file is in the archive.

Diagnostic information

.wslconfig found
	Custom kernel found: 'C:\Users\tmgfos\bzImage_6.6.4'
Detected appx version: 2.3.17.0
optional-components.txt not found
No logs.etl found in archive.
Error while parsing the logs. See action page for details

[borjamunozf]

The log size is way higher than 25Mb, 200Mb. How could we forward it to you?

[CatalinFetoiu]

@borjamunozf , if possible, can you please add the logs.etl file (and the rest of the files in the WslNetworkingLogs zip) in a one drive or google drive link and share it with us?

Thanks

[borjamunozf]

Here is it, @CatalinFetoiu

WSL Logs

let me know if you have any issue with the access.

[borjamunozf]

Any news regarding this? Anything we could do to help? Any ideas? It's really critical for a lot of devs to able to locate and fix this problem.

My best guess is that the Intel12/13th in Windows 11 with WSL is the root cause, but what do to is other story.
Thanks

[borjamunozf]

Update with more tests. It's clear that for some reason the CPU processors are not being used?

Test with disabled core isolation - memory integrity.
No result

• Test with High performance config plan and disabled core memory integry
  No result
  

• Multiple test with Process Lasso
  Setting vmmemWSL with Induced Performance Mode and the Host processor app
  Setting vmmemWSL with High performance Profile
  Setting vmmemWSL with disabled Hyperthreading CPU affinity...

This is the actual view, 11% processor usage with the tool runining

More differences regarding UDP traffic generated in Windows 10 (OK) vs Windows 11 (disastreous)
So for each packet he gets around 1516*9 + 248 = 13 892 (bits?)
And I get 1514 * 44+422 = 67038 (bits?)

• Test disabling udp packets eth0 from UDP packets of size less than 12 bytes are not sent from WSL to host #8610 (comment)
  

EDITED:

• Test with networkingMode=mirrored

• It does not work for some reason, it does not receives any data at all. Tested with firewall=true or false, applying rules but nothing.

• Test setting up distro and WSL with Windows admin
  Same result
  

Any more ideas, anything that we can try?

[borjamunozf]

More tests:

• From ROS2 documentation we establish

sysctl -w net.core.rmem_max=4194304
sysctl net.ipv4.ipfrag_time=3
sysctl net.ipv4.ipfrag_high_thresh=134217728

Testing benchmark UDP between Windows 11 to WSL2 interface with iperf

• It seems ok.

WSL2 side

iperf3 -s

Windows side:

iperf3 -c interfaceWSLIp -u -b 100MB

[borjamunozf]

MAJOR UPDATE: CC @OneBlue @CatalinFetoiu (sorry for the spam, you'll have much more work than check this thread...)

• Setting the distro to WSL1 version works fantastic performance wise!
• We're not sure if besides setting WSL1 distro is enough or any of the multiple changes we have applied.

I'm keen to consider that the HyperV is the responsible for messing this up somehow. Perhaps multicast traffic being dropped or not reaching at all.

One more note:

• lscpu in WSL1 brings correct output for core per socket (24)

• lscpu in WSL2 distro brings incorrect output (only 16 out of 24). Perhaps related to WSL 2 uses half the number of cores on AMD Threadripper 3990X #5423

We'll continue doing the tests with Ubuntu - WSL2 and try to stretch the scope of this issue and what's going on....

[borjamunozf]

Finally, it's working fine in WSL2!

Workaround:

• Use bridged networkingMode.
• Create VMSwitch External for your network adapter.

Hypothesis:

• Problem with multicast traffic with default VMSwitch used by WSL2. It seems that Internal VMswitch wont forward properly multicast given networking isolation/limitations with NAT switch.
• Theorically, I think it should work with mirrored networkinMode, but it's not.

We'll continue investigation, but it seems clearly a networking problem between multicast WSL2-Windows and not a Intel architecture one.

[Meno12ivanov]

Windows Version

10.0.22631.4037

WSL Version

2.2.24

Are you using WSL 1 or WSL 2?

• WSL 2[ ] WSL 1

Kernel Version

5.15.146

Distro Version

Ubuntu 22.04

Other Software

WSL version: 2.2.4.0 Kernel version: 5.15.153.1-2 WSLg version: 1.0.61 MSRDC version: 1.2.5326 Direct3D version: 1.611.1-81528511 DXCore version: 10.0.26091.1-240325-1447.ge-release Windows version: 10.0.22631.4037

Repro Steps

• Using CYCLONE_DDS = - rmw_cyclonedds_cpp
• sudo apt install ros-humble-ros-base

Expected Behavior

Packets are not lost in WSL & performance are the expected (~16Hz)

Actual Behavior

Three developers in my company has encountered strange performance issues in an application. This does not ocurred before without WSL!

The scenario is following:

• Windows Unity application.
• WSL 22.04 Ubuntu running ROS2
• Wireshark analyzing WSL hyper-v adapter from
  From Windows side all looks good, the topic is published every ~6ms like a clock.
  

But from WSL a lot of packages are missing. It's only receiving like 1 update per second on average.

I have established the /proc/sys/net/core/rmem_max" to higher value = 2203936 and added this to the kernel: CONFIG_IP_MULTICAST=y

It seems that this has slighly improved performance. However It starts really nice and afterwards it just get bad again.

Diagnostic Logs

No response

wsl --install
CMakePresets-schema.json

[Meno12ivanov]

CMakePresets-v3-schema.json

[sbrl]

The official website thats for mirrored mode + WSL you need 22H2 or above?

https://learn.microsoft.com/en-us/windows/wsl/networking#mirrored-mode-networking

[shixudong2020]

because WSL2(NAT)'s MTU=1420 and Windows host's MTU=1500
if WSL2(NAT) not use VPN,u can safely configure WSL2(NAT)'s MTU=1500,or configure Windows host's MTU=1420
for detail,please see WSL2 and VPN(https://blog.csdn.net/sxd2001/article/details/136788434)