Re: [Sks-devel] Optimum number of peers

[Andy Ruddock]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA512

Jeff Johnson wrote:
> 
> On Apr 23, 2011, at 2:14 PM, Andy Ruddock wrote:
> 
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA512
>>
>> Ari Trachtenberg wrote:
>>> Gossip protocols exhibit a thresholding phenomenon.  If everyone talks to 
>>> greater than a certain
>>> fraction of their peers, data will propagate to everyone in the network.  
>>> If everyone talks to less than
>>> this fraction, then very few network members will get all the data.  
>>> Unfortunately, this fraction depends
>>> on many parameters of the network ...we are working on some research that 
>>> may give some more
>>> concrete answers ... but it will take a bit.
>>>
>>
>> For me this is the important issue, it would seem that the algorithm
>> used handles having many peers extremely well - in that it would appear
>> not to lead to excessive network usage.
>>
>> I've tended to keep the number of peers I have to a small number and
>> have traditionally only peered with those who are geographically close
>> (generally speaking).
>>
>> I think it may be advantageous to have a small number of geographically
>> distant peers to prevent this "thresholding phenomenon".
>>
> 
> I suspect that avoiding the "thresholding phenomenon" isn't the right
> basis for your reasoning. Nothing wrong with your reasoning
> at all, all depends on what one chooses to optimize: in your
> case you seem to want "robust global propagation" not "minimal necessary load"
> as an optimization goal.
> 
> But that is merely a guess.

Good guess, spot on. But although I'm not seeking to minimize the load
on my servers they do perform other roles. I have to weigh up the
resources which I'm able to allocate to each service which I wish to
make available, and the fewer I have to allocate in order to achieve my
goals the better.

When I first started to operate an OpenPGP keyserver the advice was to
peer with geographically local peers - which would lead to peering wth
only a small number of hosts and would possibly lead to the "threshold
phenomenon" at continental borders (people peering within their own
continent but possibly excluding closer hosts in a neighbouring
continent). This was my thinking - with, as you say, robust global
propagation as a goal.

>> For this reason I think I shall seek to peer with one or two peers in
>> the North American continent, and one or two in Australia/New Zealand or
>> Southern/Eastern Asia.
>>
>> If this is deemed to be suitable my membership details are :
>>
>> keyserver.rainydayz.org 11370        # Andy Ruddock
>> <address@hidden> 0xEEC3AFB3
>>
> 
> For North American coverage, try peering here:
>       keys.n3npq.net  11370
> in a Tier IV data center with high availability and good bandwidth.
> 
> You are also welcome to peer with
>       keys.rpm5.org   11370
> which is an aging dual G5 on a (my) residence cable box.
> 
> Send along your membership information if interested.
> 
> 73 de Jeff

- -- 
Andy Ruddock
- ------------
address@hidden (GPG Key ID 0xEEC3AFB3)

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.14 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/
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=Q1SX
-----END PGP SIGNATURE-----