The TNY network manages a number of small servers, which we name after the periodic table elements. Trial and temporary servers are named after the unstable elements (the last ones with very strange names), and obsolete or deceased servers after the isotopes of the element that named them while in use. We name the deceased servers so it is easier to mention them without mixing up with the active servers.
All of our servers are small for good reason. We find it easier to host web services reliably when they are not at the mercy of a single point of failure. By assuming that any system can and will fail, and designing our systems to cope with this problem, we can also save in server costs. See how:
- Instead of maintaining one big server, that may be suddenly affected by a major problem, knocking our services offline for extended periods of time, we use several small ones, and distribute the load throughout them.
- We have servers on different providers and datacenters, so that when one party messes up, our services are significantly affected – we only need to temporarily put some more work on the other servers, and currently even this is done automatically.
- By planning for redundancy, we can cut costs without compromising reliability. Going for services with level agreements involving multiple-nines uptime becomes unnecessary; simple and cheap 99% uptime guarantees do.
- Data is synced across servers securely, and backed up regularly, to systems dedicated to backup and recovery. Should a major hiccup occur in the distributed systems, the impact is minor.
Resources currently in use
- Hydrogen – VPS hosted with QuadraNet, formerly Crissic Solutions before being acquired.
- Helium – VPS hosted with SecureDragon.
- Lithium – DigitalOcean droplet.
- OpenShift PaaS – used for a few auxiliary services like the service status page.