4 ms·

Sia is built on a completely different assumption than most large datacenters - uptime requirements for hosts are around 95%, which equates to about 36 hours of

by Taek 7y ago

Sia is built on a completely different assumption than most large datacenters - uptime requirements for hosts are around 95%, which equates to about 36 hours of downtime per month. This means that you can run a much leaner database and require substantially less expertise to keep things going. If a rack goes down, you don't need someone on-site to bring it up ASAP, you can fix it the next workday without upsetting your customers. 95% uptime for hosts translates to 99.99+% uptime for Sia users, because users store data across hosts in a 10-of-30 scheme. As long as 10 out of 30 of the hosts are online, you will be able to access your data. The probability of losing 20 hosts when each of your 30 hosts have 95% uptime is exceedingly small; your practical uptime depends more on the reliability of the software than it does the reliability of the hosts you use. As for commoditization: that's the goal! If we can completely commoditize data storage, prices should come down dramatically, and the market should be a lot more efficient. You won't need to be an Amazon or Google to have a stable and competitive data offering.

q3k 7y ago
> This means that you can run a much leaner database and require substantially less expertise to keep things going. If a rack goes down, you don't need someone on-site to bring it up ASAP, you can fix it the next workday without upsetting your customers. This is already the case for cluster storage systems like Ceph.
- namibj 7y ago
  This extends to the whole site. You don't need backup power. At all. Maybe no redundant network equipment, if you get replacements on speed dial. Likely no redundant fiber, just don't take your time to arrange a replacement if someone cuts it (use a different, pre-scheduled conduit you can rent and fill with fiber in a day or two). If you get unexpected "bus factor", you can handle replacement combing back early from holiday. Get creative.

by Taek 7y ago

q3k 7y ago

> This means that you can run a much leaner database and require substantially less expertise to keep things going. If a rack goes down, you don't need someone on-site to bring it up ASAP, you can fix it the next workday without upsetting your customers. This is already the case for cluster storage systems like Ceph.

namibj 7y ago
This extends to the whole site. You don't need backup power. At all. Maybe no redundant network equipment, if you get replacements on speed dial. Likely no redundant fiber, just don't take your time to arrange a replacement if someone cuts it (use a different, pre-scheduled conduit you can rent and fill with fiber in a day or two). If you get unexpected "bus factor", you can handle replacement combing back early from holiday. Get creative.