Motivation

Zebra's transaction deserialization does a lot of CPU-heavy cryptography for shielded transactions.

Since we know that transactions in our database are valid, we can deserialize them without checking the validity of curve points.

API Reference

TODO: which methods are causing this CPU usage, and what can they be replaced with?

Detailed Analysis

See #4583 (comment)

Designs

In zebra-chain:

• split out transaction deserialisation methods that optionally do validation
• do validation by default in ZcashDeserialize (with transactions and blocks from the network and RPCs)

In zebra-state:

• skip validation for transactions and blocks when reading from the state in FromBytes

Related Work

• Move database block and transaction fetches to a dedicated blocking and CPU-heavy thread #4788 (which didn't end up improving performance)