Re: [Sks-devel] Peering against "pool.sks-keyservers.net"?

[Jeffrey Johnson]

On May 2, 2012, at 1:38 PM, Kim Minh Kaplan wrote:

> Jeffrey Johnson :
> 
>> On a somewhat related note (before I dig into OCAML), can anyone
>> state the retry loop algorithm that is currently implemented for
>> non-responding peers?
> 
> It's a long time since I worked on this but if I remember correctly, the
> current algorithm is to continuously repeat:
> 
>   - wait for recon_sync_interval
>   - randomly pick a member,
>   - randomly choose one of this member's IP address,
>   - do a single recon attempt with this IP address,
> 
> That is there is no specific retry for non-responding peers. They will
> eventually get retried at random.

Perfect. That's about what I would expect for a gossip protocol like SKS.

But its not too hard to put a stopwatch on a peering operation, scale
out issues like number of keys exchanged and perhaps bandwidth,
sum withcan exponential weight to get an approx mean/stddev "cost", and then
start into adaptive "preference" rescheduling/retries (whatever that
might mean: I haven't a clue, and gossip protocols are tricky).

I suspect I can even "wing it" with OCAML sufficiently to get some
simple descriptive statistics associated with a peer however. ;-)

73 de Jeff