Build storage prototype that repurposes the LSM /DRAM spine implementation and stores columnar data in a custom rkyv friendly format which allows for 0-copy deserialization
Evaluate it by running a subset of (slow) nexmark queries
This will establish a better upper bound on what we could get from something (not rocksdb) that doesn't have the issue with alignment and isn't bogged down by column families
This will establish a better upper bound on what we could get from something (not rocksdb) that doesn't have the issue with alignment and isn't bogged down by column families