Why Relay Load Varies

Tor manages bandwidth across the entire network. It does a reasonable job for most relays. But Tor's goals are different to protocols like BitTorrent. Tor wants low-latency web pages, which requires fast connections with headroom. BitTorrent wants bulk downloads, which requires using all the bandwidth.

We're working on a new bandwidth scanner, which is easier to understand and maintain. It will have diagnostics for relays that don't get measured, and relays that have low measurements.

Why does Tor need bandwidth scanners?

Most providers tell you the maximum speed of your local connection. But Tor has users all over the world, and our users connect to one or two Guard relays at random. So we need to know how well each relay can connect to the entire world.

So even if all relay operators set their advertised bandwidth to their local connection speed, we would still need bandwidth authorities to balance the load between different parts of the Internet.

What is a normal relay load?

It's normal for most relays to be loaded at 30%-80% of their capacity. This is good for clients: an overloaded relay has high latency. (We want enough relays to so that each relay is loaded at 10%. Then Tor would be almost as fast as the wider Internet).

Sometimes, a relay is slow because its processor is slow or its connections are limited. Other times, it is the network that is slow: the relay has bad peering to most other tor relays, or is a long distance away.

Finding Out what is Limiting a Relay

Lots of things can slow down a relay. Here's how to track them down.

System Limits

  • Check RAM, CPU, and socket/file descriptor usage on your relay

Tor logs some of these when it starts. Others can be viewed using top or similar tools.

Provider Limits

  • Check the Internet peering (bandwidth, latency) from your relay's provider to other relays. Relays transiting via Comcast have been slow at times. Relays outside North America and Western Europe are usually slower.

Tor Network Limits

Relay bandwidth can be limited by a relay's own observed bandwidth, or by the directory authorities' measured bandwidth. Here's how to find out which measurement is limiting your relay:

  • Check each of the votes for your relay on consensus-health (large page), and check the median. If your relay is not marked Running by some directory authorities:
    • Does it have the wrong IPv4 or IPv6 address?
    • Is its IPv4 or IPv6 address unreachable from some networks?
    • Are there more than 2 relays on its IPv4 address?

Otherwise, check your relay's observed bandwidth and bandwidth rate (limit). Look up your relay on Metrics. Then mouse over the bandwidth heading to see the observed bandwidth and relay bandwidth rate.

Here is some more detail and some examples: Drop in consensus weight and Rampup speed of Exit relay.

How to fix it

The smallest of these figures is limiting the bandwidth allocated to the relay.

  • If it's the bandwidth rate, increase the BandwidthRate/Burst or RelayBandwidthRate/Burst in your torrc.
  • If it's the observed bandwidth, your relay won't ask for more bandwidth until it sees itself getting faster. You need to work out why it is slow.
  • If it's the median measured bandwidth, your relay looks slow from a majority of bandwidth authorities. You need to work out why they measure it slow.

Doing Your Own Relay Measurements

If your relay thinks it is slow, or the bandwidth authorities think it is slow, you can test the bandwidth yourself:

  • Run a test using tor to see how fast tor can get on your network

    For this, you need to configure a tor client to use use your relay as entry. If your relay has only Guard flag, set EntryNodes with your relay fingerprint in torrc. If your relay doesn't have Guard flag or it has Guard and Exit flags, you can't set your relay as an entry node (see https://gitlab.torproject.org/tpo/core/tor/-/issues/22204), but you can set it as your bridge, even if it is not a bridge. To set your relay as a bridge, add to your torrc:

    Bridge <ip>:<port>
    UseBridge 1

    Then download a large file using your SocksPort as a socks proxy. For this, you can use curl, eg:

    curl https://target/path --proxy socks5h://<user>:<password>@<socks-port>

    Using different user/password guarantees different circuits. You can use $RANDOM.

    That will give you some idea of how much traffic your relay can sustain.

    Alternatively, you can run relay_bw to test your relay using 2 hops circuits, in a similar way as sbws does.

  • Run a test using tor and chutney to find out how fast tor can get on your CPU. Keep increasing the data volume until the bandwidth stops increasing.