Increase expl. and i2psnark default to 3 hops

[zzz]

ref: #936 comment 4

Both will significantly increase load on network, esp. when update-via-torrent goes live. Must proceed carefully.

expl:
1) review fallback-to-shorter code
2) increase to 2 + 0-1 (0.9.8?)
3) increase to 3 (0.9.9?)

i2psnark:
1) review fallback-to-shorter code
2) interaction with update-via-torrent? load on network? 0.9.10?

[zzz]

expl. increased to 3 and expl. fallback code now allows 1 or 2 hops subtracted, 0.9.7-11 de9e9ec3f537c5ca643adffae231781ea7da47a6

[zzz]

I'm going to roll expl. back to 2 for 0.9.8 so it doesn't happen at the same time as the router rekeying. Rescheduled for 0.9.9.

[zzz]

We are nowhere near ready for 3 hop expl.

0.9.8.1-10 reduced idle time threshold from 75% to 33%. This kept us under the 50% threshold in ExploratoryPeerSelector? to pick from connected instead of not failing peers. And led to massive expl. tunnel build failure.

Will have to make EPS degrade better. We shouldn't be using all connected peers most of the time. Perhaps it can degrade by using a connected peer only for the first hop (OB) or last hop (IB).

Also, the selectHighCap condition isn't sufficient either, as when under stress the router is using that but that isn't enough. Should perhaps use connected high cap or something when under extreme stress.

The -10 change was mostly reverted in -18 to raise the threshold to 60%, or higher than the 50% test in EPS.

[zzz]

Another idea: Only increase OB first. Go to 2 + 0-1 for OB, leave IB at 2.

[zzz]

Yes. Let's do this in 4 steps, least impact first:

• OB to 2 + 0-1
• IB to 2 + 0-1
• OB to 3 + 0
• IB to 3 + 0

[zzz]

0.9.10 contains step 1
0.9.11 contains step 2
0.9.12 to be step 3
0.9.13 to be step 4

[zzz]

Step 3 was reverted in 0.9.13. It may have been a partial cause of the reliability regressions in 0.9.12, although other causes are more likely.

To be investigated further.

Still no action on i2psnark.

[zzz]

snark changed in a5566be6b23d99710db60afbf43bfd67d30d09e0 0.9.15-3