June uptime: 98.88% (down for 8h 0m)*
Uptime Tracking
The Wallpunch system is down if a valid user can’t sign up, log in, or connect. Currently I track uptime in two ways:
First, a holistic measure. A “canary” server runs a special version of the latest Wallpunch client. Once every 15 minutes the canary attempts to create a new user account, login to the system, and establish a VPN connection. If the entire process succeeds the canary pings a system uptime monitor.
(There are actually two canaries: one on a VPS in China and one outside of China. This helps differentiate between the system going down and the protocol being blocked by China’s GFW. So far no blocking in China has been detected.)
Second, a discrete measure. The Wallpunch system has a lot of components: entry nodes, exit nodes, authentication servers, and discovery portals. The uptime monitor listens for updates from each of these components. If any one component fails to check in for longer than 10 seconds the system is flagged as down.
The discrete measure has higher granularity (10 seconds vs. 15 minutes) and broader coverage (every component vs. a specific authentication path), but is less reliable. For example, a mis-configured entry node may continue to check in with the monitor even when it is unable to receive new tunnel connections.
The holistic measure would be even better if it tested the VPN tunnel after connecting to check for acceptable latency and upload/download speeds. That way the measured uptime of the system will correspond to the amount of time the system is actually working well for users.
* About 6 hours of the measured downtime in June occurred while the China canary server was down for unscheduled maintenance. I need to find more reliable canaries!
Beta Complete
Wallpunch 2.0 is a complete redesign of the Wallpunch protocol based on my experience developing the initial version. It is much more resilient, efficient, and extensible. Its key feature is the ability to combine a battery of parallel connections into a single VPN tunnel. I will explain the design in more detail in a later blog post.
The beta testing period for Wallpunch 2.0 went on a lot longer than initially planned. Most of that time was spent improving maximum tunnel throughput (now well over 300 Mbps) and eliminating protocol deadlocks. At this point the new code is quite stable and performs much better than the 1.0 version, so I’ve ended the beta and pushed the update to all clients.