Increasing thread sleeps for socket listener thread and per-connection threads

[DavidBurkett]

This significantly decreases cpu usage, especially for nodes with >20 peers. There's very little reason for the thread sleeps to be 1 ms. The only (theoretical) advantage is during mining, but as long as miners have a sufficient number of peers (~10+), there should be no practical effect.

[mcdallas]

Some results: I increased mine from 1ms to 10ms and cpu usage went from 90% to 30%

[garyyu]

I saw @mcdallas reported 10ms test result, are we sure the 5ms here has the same improvement?

[DavidBurkett]

It won't be the same improvement, but it will be close. Setting the conn thread sleep too high could have negative consequences, especially for miners. We can continue to tweak it as necessary, but this seems a happy middle ground for now.

[garyyu]

Perhaps we should merge all p2p threads into one single thread, in the future. But for the moment, it's good to have this quick and simple improvement 👍

[garyyu]

After 7 hours test, with this 5ms modification, before upgrading:

Total Peers:          112
3414 Total Connection:  116
3414 Total Established: 115

62.8%  CPU LOAD for Grin  (not stable, but a typical value)

After upgrading:

Total Peers:          164
3414 Total Connection:  164
3414 Total Established: 164

51.8% CPU LOAD for Grin (typical value)

Because the peers number different, we can't directly compare the improvement, but an estimated value 80% improvement here, anyway it deserves to have this PR, thanks @DavidBurkett

[0xmichalis]

I am seeing a ~50% cpu usage reduction in GCP - nice one @DavidBurkett !