Encryption spec

[robert3005]

There's still some extraneous comments in the flatbuffer that need answering

[a10y]

I think for 2MB segments the performance difference could be meaningful for random-access. Basically if we use GCM and we have multi-MB blocks, we lose the benefit of the granular zone maps. Maybe that's just the price you pay?

[robert3005]

I think we can write a benchmark to measure this. Would tell us how big is the difference. There’s also storage overhead. Generally nonces in gcm are random but ctr lets you get away with counter

[robert3005]

I think if we want to add ctr mode we benefit from additional storage place for IV somewhere next to segment spec so we can actually randomly seek without reading first few bytes this is irrelevant whole segment is already in memory

[gatesn]

this is irrelevant whole segment is already in memory

This isn't necessarily true for memory-mapped files

[robert3005]

I think for memory mapped files the overhead of reading additional 12-16 bytes might not be that relevant but could would let us skip loading one extra page