Protocol traffic

[jazzor]

Since 1.10RCx i noticed theres a new protocol traffic measurement, which is nice to see, but i cant help but question its accuracy/my settings because its using up A LOT of the bandwith. My upload is set to 16kb/s and 1 running torrent and the traffic is a whopping 8kb/s. What causes protocol traffic to go so high?? Is the meter wrong?? Less connections??

[andar]

Since 1.10RCx i noticed theres a new protocol traffic measurement, which is nice to see, but i cant help but question its accuracy/my settings because its using up A LOT of the bandwith. My upload is set to 16kb/s and 1 running torrent and the traffic is a whopping 8kb/s. What causes protocol traffic to go so high?? Is the meter wrong?? Less connections??

It's proportional to the rate at which you are downloading/uploading.. For instance, when I am downloading at ~1.2MiB/s, my upstream protocol rate is close to 50Kib/s. Reducing the number of connections will help a bit because there will generally be less protocol chatter.

The way TCP/IP works is that when you receive a packet from a peer, you must send an acknowledge packet back, which is also at least 40 bytes (the protocol header size) and this is added to your upstream rate. So let's just assume you are receiving packets in the size of 1500 bytes and responding with a 40 byte packet, that means that at a minimum you need to 2.7% (40/1500) of your download rate back in your upstream just to account for ACKs alone. This does not account for bt protocol chatter to peers requesting pieces, etc..

[jxx]

How can the protocol overhead be affected by the number of connections?
If I send an ACK to peerX or peerY the packet size is the same, isn't it?

Just wondering...

[andar]

How can the protocol overhead be affected by the number of connections?
If I send an ACK to peerX or peerY the packet size is the same, isn't it?

Just wondering...

The more connections, the more ACKs you will be sending.

[jxx]

I just noticed that the upload overhead is normally a lot higher than the download overhead...

Example from a few hours ago:
D/L 190KiB/s with a protocol traffic of around 4 KiB/s
U/L 25KiB/s with a protocol traffic of around 8 KiB/s

And my current readings:
D/L 30KiB/s with a protocol traffic of around 2 KiB/s
U/L 20KiB/s with a protocol traffic of around 3.5 KiB/s

Not really proportional, and not really logical either, imo.

Still wondering...