Mirrored

Alternative Hosts

Mirrors usually describe some ‘place’ where a copy of some information is stored and accessible. In contrast, a proper backup intended for longtime archiving and recovery purposes would not be included in this category as they are typically not accessible in the same way as the master copy.

Mirrors of software repositories are often found within organizational networks, for example, corporate intranets. For instance, products like Nexus and Artifactory allow the creation of self-hosted repositories. Besides hosting custom software packages, a mirror or proxy feature enables the integration of upstream repositories, for example, public systems like the NPM Registry. They seamlessly combine custom and upstream artifacts into a single repo, typically consumed within a given organization. One has to mention that the mirror feature does not initially copy all content from an upstream system but downloads artifacts on demand and stores them for re-distribution. Also, it is possible to make such a system publicly available to potential customers, rendering it a one-stop shop for all their software needs.

Typical use cases for such local mirrors include:

  • Reducing network load on the upstream repository and the local internet connection. An artifact is downloaded from the internet once and subsequently only transferred within the local network.
  • A central, localized copy of artifacts reduces latency and increases transfer speeds, as internal networks are usually faster.
  • A central copy that is quickly and easily accessible removes the need for other sophisticated caching mechanisms.
  • A local mirror might have a higher availability or, at least, is unaffected by an upstream repository’s downtime.
  • Additional security scans can be performed on the synchronized data.
  • An organization can detect and prevent upstream content manipulation in the local copy.

Generally, a mirror system does not have to copy all the content from an upstream repository. Instead, the operator actively excludes certain artifacts or includes additional content to fulfill the system’s purpose.

Mirror Networks

Architecture

Large-scale networks of repository mirrors are an alternative setup to the centralized, cloud-based approach discussed in the previous chapter. Notable examples are the Arch Linux Repositories and the Debian Repositories.

Such systems are based on a Master-Slave-Design similar to Database Replication. Here, a single master repository is owned and operated by the core teams. Their members and selected volunteers can contribute new content to the repository. The master generally allows read-and-write operations.

Mirror repositories, however, are read-only replicas or slaves. This means that the content from the master repository is synchronized to each mirror in defined intervals. This is a one-way operation. The mirror operators are generally not able to modify the content. Users are only able to obtain artifacts from the mirror repositories. If they want to contribute artifacts, they must upload them to the master and await the eventual synchronization to all mirrors.

Mirror Network

Mirror Network: Mirrors worldwide synchronize the content from the master repository and serve the data to users in their vicinity.

A multitude of these mirrors is hosted by volunteers around the world. These are often corporations, ISPs, and universities that possess many unutilized resources like, for instance, bandwidth and storage space. Therefore, a regular user should prefer a mirror to the master repository to download artifacts, as the master might be overloaded with numerous requests. Furthermore, a mirror is most likely located closer, enabling lower latency and possibly higher transfer speeds, notably if the mirror is situated within the same network segment, e.g., within the ISP’s network.

The traffic load is spread around the network. This reduces the costs for the master as it does not have to scale up to meet the entire community’s needs. As a result, a highly productive network is created as a community effort, which induces little cost to individual parties. Additionally, the performance and overall experience for users are improved.

A Bit Less Centralized

Involving independent volunteers to host mirrors creates a less centralized network. The mirrors themselves operate autonomously. If the master is unavailable, the mirrors continue to serve data based on the last synchronization.

However, the master is still in absolute control. They could block users and mirrors from downloading artifacts, and they could modify the content at their will, which is subsequently synchronized throughout the network. Nevertheless, mirror operators are in control of their respective systems. They could similarly block users and stop synchronizing with the upstream system if they disagree with the content.

From a user’s perspective, this sounds bad at first. But due to the sheer amount of mirror operators, chances are that there are at least a few who will serve you up-to-date content even if you are the worst supervillain.

However, one problem persists. The master controls the content of the entire network as they are the only ones with write permission.1 In theory, mirror operators could change the content. But due to checksums and digital signatures effectively spanning the entirety of the content2, changing the content without the master’s private keys would result in an invalid state. Clients check for the validity of the content using these known keys and decline packages that fail this verification. The only measures a mirror operator can resort to are stopping synchronizing or effectively forking the content.


  1. In truth, it is actually not much of a permission but the system design that every mirror copies the data from the master. This makes it effectively the only writing instance in the network. ↩︎

  2. Artifacts in the repository are digitally signed by the core teams, who upload data into the master repository. The indices, basically the file tree structure, are signed similarly. Consequently, the state of the entire content is digitally signed, which makes manipulating it practically impossible. ↩︎

7 / 18