L1 cache miss escapes to the X closest peers L2 peers in the L1s Swarm

[aarshkshah1992]

On an L1 cache miss:

• Given a key K, the L1 node finds the X closest L2 peers in it’s Swarm to K using the XOR distance metric. X is the replication factor of the L2 Swarm and is picked such that not all X nodes go offline between two requests to the Swarm for the same content. We can start with a low value for X(say X = 3) to reduce the amount of data we need to move around in the MVP and then tune X based on empirical data post MVP.

• The L1 executes the cache miss by sending concurrent WANTHAVE messages to all the eligible L2s.

• It then picks the first L2 that replies as the L2 responsible for serving the cache miss and continues the transfer with it.

• It sends a CANCEL message to all other L2s.

• If all L2s come back with a negative/don't have response, we defer to L1 should cache miss directly to the IPFS Gateway for key K if all the X closest L2 nodes do not have it #5 AND L2s should follow the cache on second miss rule and hydrate their persistent cache asynchronously from IPFS Gateway  #6.

[aarshkshah1992]

This needs more sub-issues for better granularity and for breaking down the work into more sub tasks.