Description
What is the problem?
Profiling shows the majority of the CPU time increase when taking an encrypted backup is from rand.Read when generating a nonce. If we improve this the throughput should improve
What is the solution?
A couple of ideas:
- Rather than generating a single nonce at a time we could generate a bunch and advance through the list. Once there were none left we could synchronously generate more
- Generate nonces asynchronously on another thread. Allow a buffer so ideally there is always one ready
Attachments
Gerrit Reviews
For Gerrit Dashboard: MB-62043 | ||||||
---|---|---|---|---|---|---|
# | Subject | Branch | Project | Status | CR | V |
210457,2 | MB-62043 Generate nonces on a separate thread | master | backup | Status: ABANDONED | 0 | -1 |