...
- How much metadata space is needed for all tables/applications?
- Up to 2^48 objects, 128 bit (16 byte) key, size (4-8 bytes), permissions or appid if not in address (2-4 bytes)
- (2^43)*16 = 128 TB
- How does metadata replication occur and what is the frequency?
- Lazily
- Not sufficient when a client discovers a host is down
- must update mappings in the new replicas at least very quickly
- May additionally want leases or heartbeat or something similar as in MapReduce to ensure enough copies of shards are maintained on failure even if the data is cold
Approaches
...
addr mod servers
- + Balanced storage
- Statistically, if keys are random
- Need other protections if keys are not random (e.g. user can control placement)
- - Requires same amount of storage in all machines
- - Adding/removing machines
- Requires all objects to move, or for cluster to only double in size each time
Mapping/Address Space Partitioning
- + Adding/removing easier
- Layer of indirection shard -> host
- - Requires reverse lookup addr -> shard, requires a range search
RAMCloud Address -> Metadata as quickly as possible (e.g. with as few requests and as little processing as possible) where metadata includes physical address, size, and permissions at least.
...