[INFO] TCP Fragmentation Test

[MDA2AV]

Is not allowed to be ran throught actions on github, setting the MTU to 69 also causes some modifications to kernel config we can't reverse yet, those changes reduce slightly the performance overall for other tests.

The tcp-frag test will be ran manually by me when required.

[remittor]

It is likely that the behavior of different servers differs when receiving Noise requests.
Because of this, we see different results.

@MDA2AV , What should the HTTP server respond to Noise requests?

[MDA2AV]

It should respond a 4xx for noise, some do respond 2xx because they have a fallback to unknown routes, a validation for this will be added

[joanhey]

The important question, is that no framework maintainer can test isolated each of these tests !!
Without that information, it's impossible to improve the results.
The bench is for final users, but it's more important than the maintainers have the information to improve.

[MDA2AV]

The important question, is that no framework maintainer can test isolated each of these tests !! Without that information, it's impossible to improve the results. The bench is for final users, but it's more important than the maintainers have the information to improve.

Users will be able to test locally using benchmark scripts.

[joanhey]

First we can test using the benchmark script locally, #390 (comment)

And later we need to check that work for that individual test from the mixed.

[joanhey]

The information of this bench is more important for the maintainers, to improve the code, than for the final users.

The final users will check different results numbers with any round.

[joanhey]

@remittor we test it also the noise request and is working OK.
And in the bench show thousands of errors !!

[joanhey]

After checking the test for me, this endpoint don't exist, so need to return a 4xx response error for the noise test.

How can we test this endpoint ??

[MDA2AV]

Is not allowed to be ran throught actions on github, setting the MTU to 69 also causes some modifications to kernel config we can't reverse yet, those changes reduce slightly the performance overall for other tests.

The tcp-frag test will be ran manually by me when required.

Did you read this????!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

We disable this test to run on benchmark script protect people from screwing up their kernel configs!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[joanhey]

So we can't test it, so how we can improve it ?

[MDA2AV]

Noise endpoint IS supposed to return a 4xx.

If you want to improve TCP fragmentation handling in your framework set the MTU on your own and run normal load generator. But we won't have the script modify Kernel configs in your machine, that can break people's system.

you can pass --frag flag to benchmark.sh at your own risk! That allows running tcp-frag test

If anything happens to your system, we are not responsible.

[joanhey]

No standard networking protocol or common device uses a TCP MTU of 69 bytes, and the search context does not reference this specific value.

Standard Ethernet networks typically use a default MTU of 1500 bytes, while older WAN links with significant overhead (such as PPPoE) often require adjustments to 1492 bytes or lower to avoid fragmentation. The Minimum MTU specified in IPv4 standards is 576 bytes, which is the smallest packet size that all IP implementations must support, but a value of 69 is not a recognized standard configuration for TCP or IP traffic.

So why exist this test ??