Why is my measured bandwidth so low compared to actual — is there a off by 1K issue?

Question

I've noticed my Tor relay measured bandwidth is very low compared to both my advertized and external-to-Tor measured bandwidth. Based on the following observations, I wonder: is there's a 1K scale factor issue in the Tor measured bandwidth processing?

I'm running a dedicated Raspberry Pi 2 node with up-to-date Raspbian and wired ethernet; I pulled the vote and consensus data from https://collector.torproject.org/recent/relay-descriptors/.

Units: based on biology experiments, I conclude that the bandwidth votes are measured in kB/s, yet bandwidth is treated like B/s (at least in arm).

Advertized in votes: "1024"; therefore I conclude vote units are (kB/s), since I proposed 1000 KB in my torrc.

Measured in votes: We had the following for measured bandwidth: { n/a 1680 463 n/a n/a 658 n/a 191 n/a }

The resulting consensus: 463 (reasonable outcome given 1680, 658, 463, 191)

The reported measured bandwidth (from arm): 3.6Kb/s

The math (note that is 463, not 463e3 nor 463*2^10): 463B/s * 8b/B / 1024b/Kib = 3.617Kib/s

So it sure looks like the bandwidth votes are in kB/s, while it is treated like B/s when used from the consensus.

I suspect this low measured bandwidth is the basic reason for my low weighted mean middle score (~5e-5), and hence low consumed bandwidth (~90kB/s).

================================================================

FYI, here's my observed bandwidth available into/outof my Tor relay -- my Tor relay is plenty fast, compared to Tor network measured and consumed.

    host    download                upload
    relay   30.1 Mbps 3.8 MB/s      11.5 Mbps 1.4 MB/s

See also Arm for Tor Relay Bandwidth Descrepancies

Answer 1

Thanks for running a relay. (:

Clients choose which relays to use based on their consensus weight. The consensus weight is assigned by the directory authorities after measuring the throughput for your relay.

There are a number of things that can impact the speed at which your relay can be used, including the CPU speed, CPU architecture, the network interface and the network environment around you.

It may be that you are just unlucky and the bandwidth authorities have a poor connection to you, while you may have a really good connection to other sites.

The bandwidth authorities will also group you with similarly rated relays when they perform the measurement, this means that your maximum measured speed is limited by the slowest relay in the group. You may have had a poor measurement when first measured, and are now limited by the group you've been placed in.

I wouldn't worry about the speed too much. I see currently it's 3.17MiB/s (which is 25.36Mbps), which is a very respectable speed for a relay.