Benchmark HaloDb as a replacement for RocksDb

[GalRogozinski]

Description

Benchmark HaloDB (https://github.com/yahoo/HaloDB) as a possible replacement for RocksDb
Find out if we can get a better read throughput without paying too much in write throughput

Motivation

Currently we are not happy with the high cost of RocksDb get operation. We are thinking of implementing our own cache (see #935). Perhaps using a db that that is native to java can save us from doing that.

Requirements

1. Compare read throughput with RocksDb
2. Compare write throughput
3. Compare resulting DB size

[StephanSalas]

I'm glad you thought of this, and it's always good to improve efficiency. That aside, I thought of 3 main considerations:

1. How maintainable a lesser-known, or custom solution would be in replacing an already-working database?
2. I read through Higher level cache above the standard RocksDB cache #935, but was missing a concept of the "What is good enough?" question. It's always good to improve efficiency, but once again how maintainable would another cache abstraction be and is the benefit worth the effort?
3. What's our approximate efficiency net benefit of this/these change(s)? (This'll probably need more research, as you mentioned).

Not to beat a dead horse with 3, but yeah just some things to consider.

Best,
Ss

[GalRogozinski]

Thanks Stephan!
We will answer your points before closing the issue