The Design Problem ¶
In a centralized world, users trust a particular server with data. It’s assumed that the website will always be online, and if it goes down, you aren’t able to access that data.
In a decentralized world, we’re able to increase the resilience of data by relying on multiple providers instead of a single provider. This means that if for some reason one service loses your data, another provider will have a copy.
However, in a peer to peer or federated application, each of these servers is run by a different entity. It’s sometimes unclear how and when these servers are online.
The Design Solution ¶
Monitor and record device that downloads data. Keep a history of the time that data was synchronized for each device, and periodically check on these devices to ensure they have the latest copies. You can use this in conjunction with conditional-sharing to intelligently synchronize data to new devices.
Allow the user to see what other devices have access and are re-hosting their data to the network. Visualize this information in the user interface at multiple scales depending on the details necessary per screen. You can use this in conjunction with age-indicator to understand how long it’s been since a device has seen another, helping users understand if their data is safely replicated to another device and they can turn off their computer.
Why Choose Network Heath Indicator? ¶
- Confirm that shared data remains widely available
- Confirm that hosts supply data for a long time
- Confirm that hosts do not continue providing files after a deletion request
- Identify peers with similar interests
Best Practice: How to Implement Network Health Indicator ¶
- Allows users to perform self-diagnosis and debug their own network and get updated on the uptime of other devices.
- Provides a variety of different network indicators such as confirmation of uptime, number of active connections, percent downloaded, etc.
Potential Problems with Network Health Indicator ¶
There can be a lot of information about each device that isn’t really useful to all users. Some users will want to see advanced information, like the IP address. Consider ‘advanced’ and ‘basic’ views that users can toggle on or off depending on what they need from the interface.
Keeping a local database may not be enough to have the full scope of history, especially in peer to peer applications. Consider gossiping the data as part of the replication protocol. For example, if Bob synchronizes with Sally, and then logs off. Later Sally synchronizes with John, and John logs off. When Bob logs back on, he will not know that John also has the data. Sally’s device should automatically tell Bob that she saw John while Bob was offline. This will ensure that users know who has seen the latest information.
Some protocols by default will not have the ability to acknowledge or verify how much of a dataset has been replicated by particular devices. This is required to make Network Health Indicator more informative and accurate.
Network Health Indicators may be unreliable in offline (i.e., sneakernet) networks, where people share data offline using hard drives or other physical methods. This is also true for applications where users may disconnect from the network for large stretches of time. Implementing indicators in these environments may require tracking detailed network health history rather than continued uptime. Alternatively, utilize a notification protocol, such that peers notify one another when they are connected and available, rather than nodes polling all peers periodically to discover availability.
The Take-Away ¶
Network health indicators reassure users and build trust in your application.
References & Where to Learn More ¶
Network health indicators overlap with reputation and trust management, in that hosting data for a long period of time can be used to gauge the reliability of a peer. See cautious optimism and conditional file sharing. There are potential applications related to preventing DDoS and Sybil attacks.
Tracking peers that re-host content can be used for social peer discovery, because someone mirroring files on an obscure topic likely shares that obscure interest.