Question: performance figures

[psantosl]

Hi,

Do you have any performance figures comparing libutp to TCP?

I've found this https://maidsafe.atlassian.net/browse/MAID-3221?page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel and the results mostly match what we've got running ucat ourselves, libutp being about half fast as TCP. But I'm shocked by the results, I expected libutp to be much faster, so I bet ucat is not valid for a real benchmark, am I mistaken?

[Seeker2]

Often libtorrent-based uTP struggles to reach half its TCP speeds, so yeah...those are valid results.

[anacrolix]

I maintain a Go torrent client that uses libutp extensively. I can anecdotally report from my own experience, and that of others that uTP does not necessarily provide the same throughput that TCP can. Of course there are other benefits and reasons for this which are reasonable.

[the8472]

Is testing via loopback a realistic though? ~0 latency, very high bandwidth and packet losses that don't reflect the behavior of physical links. I guess it stresses different parts compared to WAN connections for which utp is intended.

That said, those speeds still seem low for such an idealized connection.

[Seeker2]

@the8472 I've tested with someone a couple continents away with some built-in packet loss and very high latency. I've tested with ordinary torrents to people on similar ISPs as myself. I've even tested on other computers and 2 different ISPs in my own house. uTP performance on libtorrent (when using Windows!) is not on par with TCP speeds in ANY of those tests.

I resorted to using mostly (but not always) 127.0.0.1 local loopback because it's easy and quick to set up compared to the coordination needed to do the others.