Help needed diagnosing slower than expected throughput

[rageshkrishna]

Hi folks! I'm running 2 libtorrent peers on Digital Ocean. They are not part of the mainline bittorrent network. They're only serving some of my own torrents and using my own tracker to find each other. The behaviour I'm seeing is that the throughput between these nodes seems to be maxing out at 100 - 130 MB/s. I know for sure that the nodes can do 200+MB/s between them (verified with iperf3 and also python's http.server). If I try a single torrent, it gives me around 130MB/s but if I try two torrents in parallel, they are not able to push beyond 130; both of them share this transfer rate equally, giving me something like 55 MB/s each.

Here is my setup:

• Using SSL torrents
• Only TCP is enabled. UTP in/out are disabled.
• upnp, natpmp are disabled
• DHT is disabled
• allow_multiple_connections_per_ip is enabled
• Active seeds, downloads, limit are all set to -1
• Tracker is external, so nodes are announcing their public IPs even though they are on an internal network. Even so, iperf3 shows me that the DO machines can do 200+ MB/s even with the public IPs so there is bandwidth to spare.

Using this, I test with two torrents. The first is 1484 MB, and this gets downloaded in around 10 - 12s on its own. The second is 1659 MB and this too is somewhere in the neighbourhood of 10 - 15s on its own. If I try both together, they tend to take 25 - 30s, which means throughput is only around 100 - 120 MB/s.

What I've tried so far:

• Set almost all of the high_performance_seed settings, except leaving active seeds, downloads, limit at -1. I have only around 10 torrents I'm testing with so unlimited active shouldn't really be a problem
• Tried bumping up max_out_request_queue and max_allowed_in_request_queue to insanely high levels. This gets rid of the performance alert, but doesn't have a significant impact on the bandwidth used. In fact, I think that setting extremely high levels are making the whole thing a little slower (maybe by a few seconds).
• Captured session stats at the end of each download and plotted the graphs, but I'm not sure how to interpret this data.

The net effect of this is that I am able to download the same data from the internet faster than I am able to transfer them between these peers on the local network, which is the complete opposite of what I had expected. Downloading from the internet, I'm easily able to get 80 - 100 MB/s on both downloads in parallel, so it's a lot faster than the 55 MB/s that each torrent is giving me in parallel.

My questions:

• Is my expectation that a single seed can serve 2 torrents at 200+ MB/s to a single peer on the local network valid?
• Is there any guidance on how to interpret the session stats output? What should I be looking for?
• Are there any other performance guides or tweaks that you can recommend?
• Is there something else I can do to identify my bottleneck?

Thanks in advance!
-- Ragesh.

[rageshkrishna]

I ran a few tests and found that the entire perf degradation comes from using SSL torrents. Turning off SSL torrents gets it to use 100% bandwidth with no special configuration.