performance regression in v3.4-ose

[mbiebl]

Hi,

I noticed a significant performance regression in newer versions.
My test setup is the sync of an address book with 1000 contacts, test device is a Pixel 4a.

With v3.3.12-ose a complete sync takes about 40s, with v3.4-ose it takes roughly 1min 30s, so more then twice as much.

A git bisect shows 450368b as the first "faulty" commit.

commit 450368bd0ef248e483e250318800cf29d86531d9
Author: Ricki Hirner <[email protected]>
Date:   Mon Aug 23 11:05:36 2021 +0200

    Use ThreadPoolExecutor with dynamic number of threads as global sync work dispatcher

Any ideas, thoughts?

[rfc2822]

Possibly the number of threads. It depends on how much load we want to put on various devices… for instance, you can synchronize with all CPUs and 32 concurrent connections, then it will take less time but the device may not be as reactive. The other side would be to use little CPU, but then more battery may be needed because the (mobile) connection must be hold for longer time.

If you have a suggestion for a good algorithm to balance speed, load, and battery, please post it here :) In (hopefully) 4.4 the sync algorithm will be rewritten so that would be a good time to incorporate such thoughts.