0

I am trying to develop a real time data streaming application with different size of messages using tcp. In fault injection tests it is observed that when the network delay is less than 150 ms, messages of 100 bytes are much easier to be lost than the ones of 500 bytes. However under high network delay the loss rate of large messages(more than 500 bytes) increases to over 0.3% while the loss rate of smaller messages still remains at %0.1. Can anyone provide some idea why this is happened? Thank you in advance.

1 Answer 1

1

it is observed that when the network delay is less than 150 ms, messages of 100 bytes are much easier to be lost than the ones of 500 bytes

That is an artifact of your simulation.

Generally, delay and loss ratio have no relation. Packet loss has two main reasons: transmission faults and congestion.

The transmission fault probability is (usually) higher for larger packets since they spend more time in transit. It may also be higher for connections with longer delay for the same reason.

Loss by congestion depends on the actual devices used and their queueing logic, and is largely independent on delays. However, longer delays often mean more hops (active processing steps), so a higher chance of running into congestion.

3
  • Thank you for your pure explanation and I have one more confusing question according to me. When i test relationship between packet loss and message loss rate until packet loss rate increasing to %10 there is no loss in messages. However when i set packet loss %10 then i get %20 message loss rate. Do you have any idea or this is again artifact of my simulation. Thank you again for your reply. May 2, 2020 at 13:43
  • not %20 it should be %0.2 May 2, 2020 at 13:52
  • TCP can only recompensate so much packet loss. If repeated retransmissions fail to reach the destination the connection will eventually time out. 10-20% packet loss are horrendous, like a barely (or rather not) working Wi-Fi. Packet loss in wired LAN should be less than .0001% (1 ppm). Even for Internet links .1% is a lot, and 1% is inexceptable. I think you'll need to check the specifics of your simulation, it seems to produce a lot of artifacts.
    – Zac67
    May 2, 2020 at 16:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.