Speed and Resiliency through Forward Error Correction

What we do, replicate :slight_smile: So there are 8 copies of each chunk all geographically distributed due to xor space etc.

At the moment
You request a chunk, 7 nodes give you a small message with hash of chunk and who is sending it. The sender node sends the data.

Better

All 8 send small message with hash of data. You select the first one to answer and get the whole chunk from it. Then you get the data at fastest network speed but at cost of an extra message.

5 Likes