Skip to main content

Impact on Trackers

Automatic cross-seeding is a relatively new practice, and while it brings some benefits for trackers, it also has some drawbacks. In this doc, we will outline the benefits and drawbacks, then discuss strategies you can use to minimize cross-seed's footprint.

Benefits​

The benefits a tracker sees from automatic cross-seeding largely center around retention. Having more seeders leads to healthier swarms, longer retention, and fewer dead torrents.

It is particularly helpful for new trackers looking to build a catalog—often, backfill uploads are not particularly well-seeded because the users have already downloaded them from elsewhere. But any cross-seed users will automatically pick up the new releases and reliably seed them, increasing redundancy.

Drawbacks​

Private BitTorrent trackers are small-budget hobby software projects that run on donations and the good will of the sysop. They do not run on infrastructure like Google's or Amazon's, and they cannot support infinite load.

Automatic cross-seeding causes more load on trackers than any previous form of automation (e.g. Sonarr). We, and the tracker administrators we've spoken to, believe the benefits outweigh the drawbacks. However, not all trackers hold the same opinions, and we haven't spoken to all of them.

What is "load"?

We can break load into 4 distinct actions that cross-seed (and all other tracker automation) performs:

RSS requests​

  • These are performed by rssCadence and cross-seed rss.
  • cross-seed initiates an RSS request by searching with an empty query. These are usually cached and updated regularly, so the tracker's server is sending you a precomputed file.
  • These are cheap in both compute and network I/O.

Search requests​

  • These are performed by searchCadence, cross-seed search, and webhook searches.
  • cross-seed searches for movie and TV show names individually.
  • The tracker's server has to scan through its catalog looking for releases with similar names. Using id-searching not only yields far better results but is also a lighter load on the tracker database.
  • These are expensive in compute, but cheap in network I/O.

Announce requests​

  • These are performed by Announce Matching.
  • A tool such as autobrr uses IRC to listen for announces from the tracker and then sends them to cross-seed.
  • This has zero load on the tracker since they create the announce once and publish it in the IRC channel for everyone to see, regardless of the number of users. Only the snatches from announce create load on the tracker, if it passes cross-seed's pre-snatch checks.
  • These are completely free in network I/O and compute.

Snatches​

  • cross-seed will snatch any torrent that has the same release group, source, resolution, and rough size as your owned torrent.
  • The tracker's server has to add your passkey to its stored copy of the torrent file, then send it to you (and it's often quite big—a few megabytes).
  • These are expensive in network I/O (and maybe memory), but cheap in compute.
  • Some trackers count snatches for their user rank systems, and snatching many unnecessary torrents may be seen as abuse.

How cross-seed Minimizes Load on Trackers​

cross-seed is designed with these drawbacks in mind. While it's a tool for users, it should be as friendly as possible to trackers.

Category-aware searching​

cross-seed queries each indexer for the media Torznab categories it supports, and skip searching for any media type the indexer doesn't carry. This heavily reduces "useless" searches and makes it more feasible to use cross-seed with non-video trackers.

Preventing repetition​

Similar search requests are grouped together in a run and share the results if applicable. This prevents duplicate queries for searchees that differ only by release group, source, resolution, etc. These results are not cached long term as they can change over time.

cross-seed also caches the snatches it makes. This means on subsequent searches, cross-seed will never snatch the same torrent twice—the biggest load on a tracker.

Parsing bad titles​

cross-seed will parse bad titles to add aditional information to the search query. For example, a torrent titled "Season 1" could be parsed to "Show S1" if the information is available within its files. Similarly for data-based searches, cross-seed can parse the standard Sonarr and Radarr folder structures for more meaningful searches. It will also outright reject searches that cannot be salvaged.

How cross-seed's Default Settings Minimize Load on Trackers​

For applicable options, cross-seed limits the values to prevent badly configured setups which harms both the user and the trackers. The default configuration also aims to minimize load by being selective about which torrents it searches, and by working through its backlog slowly.

Delay between searches​

As of version 6, cross-seed enforces a minimum 30 second delay between individual search requests during any bulk search. This only applies to bulk searches, not between each webhook-triggered search.

Preventing unnecessary searches​

cross-seed enforces that you have configured the excludeRecentSearch and excludeOlder options, and its defaults are set such that cross-seed will search for things recently downloaded a few times, but only search once for things downloaded far in the past. These settings will build your backlog of cross-seeds, and sufficiently catch new releases trickling to other trackers.

Skipping episodes​

It's usually not worth it to bother cross-seeding episodes since they will be deleted in the near future (when season packs come out). By default, cross-seed skips episode torrents for search and rss with includeSingleEpisodes: false. However for Webhook Triggered Searches and Announce Matching, cross-seed treats these differently since these are triggered on a new download or upload, meaning a season pack is unlikely to be available.

Similarly, cross-seed will not search individual episodes from a season pack as these are likely trumped/dead.

If you are racing or never download season packs, and would like to turn on episode searches, set includeSingleEpisodes: true.

caution

There are TV shows that are not combined into season packs, such as game shows, talk shows, and similar. These shows are known as dailies and often have naming which specifies dates, rather than SeasonEpisode (S02E03) style. Searching with includeSingleEpisodes and using flexible or parital matchMode on these torrents, or files, can result in many torrent files being downloaded for comparisons for matching only one individual episode.

Remember, this would occur on just one (torrent) search, within the entire (all data) search.

Skipping torrents with significant unknown/extra files​

This behavior is covered in detail here.

Limiting search volume​

By default, cross-seed only make 400 search requests from your backlog per day. This is measured by the number of unique search queries that cross-seed makes per run. We do this for a few reasons:

  • It eases you into the process of cross seeding. This limit is only consequential when first starting out.
  • If you have a high excludeRecentSearch of one year for example, intending to do a rolling daily search of the torrents that were last searched a year ago, torrent searches will "bunch up" once a year (depending on when you first set up cross-seed) to be searched back to back, potentially causing rate limiting problems for your other automations due to the high search volume (Sonarr, Radarr, etc.).
  • Long-lived bulk search runs encounter drift - while they run, your other automations will be downloading new torrents, moving data files around, and deleting torrents and/or data files, and the snapshot cross-seed took of your torrents at the beginning of the search goes out of date. Keeping bulk search runs short helps a lot.

You may have an instant gratification mindset, and we get it! Seeing your total number of cross-seeds and your final seed size on a tracker is exciting. With that said, please try to balance that mindset with being friendly to your trackers.

The important thing is that you seed for a long time - not that you start seeding as early as possible.

This is configurable with searchLimit.