Optimum number of gossip peers; mailsync with PKS network

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Optimum number of gossip peers; mailsync with PKS network

Jonathan Oxer-4
Thanks to all the peering offers I now have 6 gossip partners for
keys.keysigning.org, which got me thinking about the optimal number in a
mesh network. There's obviously overhead in the sync protocol and it's
wasteful to check more frequently than there are changes to propagate,
so the theoretical optimum check frequency would vary inversely with the
number of peering partners and proportionally with the rate of changes.

Short version: how many peering partners is normal?

Also, now that SKS gossip is working I want to set up mailsync to the
PKS network. The 2004 docs on the wiki said that permission should be
requested before adding a PKS server to the mailsync file, but then says
that [hidden email] can be added without requesting
permission.

Is there more up-to-date information about PKS sync? What's the current
best approach?

Cheers   :-)
--
Jonathan Oxer
Ph +61 3 9723 9399

_______________________________________________
Sks-devel mailing list
[hidden email]
http://lists.nongnu.org/mailman/listinfo/sks-devel

signature.asc (196 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Optimum number of gossip peers; mailsync with PKS network

Yaron Minsky-2
I think the optimal thing would simply to have everyone in your peer list.  I don't think you lose anything to that, but it's a pain to maintain.  I think there are some theoretical results suggesting that one can do with O(log n) peers where n is the size of the network, assuming the peers are chosen at random.

In practice, gossip is pretty damn robust.  I generally think 5-10 peers is about right, and one shouldn't need to change that too much as the system grows.  For reasonable values of n, log n is constant.

I like a sync frequency of roughly a minute.  That means the overall network distributes updates pretty quickly -- basically you should expect an update to have fully percolated within 10 minutes.  Since each gossip reconciles -all- differences, you don't need to change the gossip rate as the update rate changes.  And updating once  a minute is pretty light in terms of resource usage.

y

On 10/29/07, Jonathan Oxer <[hidden email]> wrote:
Thanks to all the peering offers I now have 6 gossip partners for
keys.keysigning.org, which got me thinking about the optimal number in a
mesh network. There's obviously overhead in the sync protocol and it's
wasteful to check more frequently than there are changes to propagate,
so the theoretical optimum check frequency would vary inversely with the
number of peering partners and proportionally with the rate of changes.

Short version: how many peering partners is normal?

Also, now that SKS gossip is working I want to set up mailsync to the
PKS network. The 2004 docs on the wiki said that permission should be
requested before adding a PKS server to the mailsync file, but then says
that [hidden email] can be added without requesting
permission.

Is there more up-to-date information about PKS sync? What's the current
best approach?

Cheers   :-)
--
Jonathan Oxer
Ph +61 3 9723 9399

_______________________________________________
Sks-devel mailing list
[hidden email]
http://lists.nongnu.org/mailman/listinfo/sks-devel




_______________________________________________
Sks-devel mailing list
[hidden email]
http://lists.nongnu.org/mailman/listinfo/sks-devel
Reply | Threaded
Open this post in threaded view
|

Re: Optimum number of gossip peers; mailsync with PKS network

Jason Harris
In reply to this post by Jonathan Oxer-4
On Mon, Oct 29, 2007 at 10:54:20PM +1100, Jonathan Oxer wrote:

> Also, now that SKS gossip is working I want to set up mailsync to the
> PKS network. The 2004 docs on the wiki said that permission should be
> requested before adding a PKS server to the mailsync file, but then says
> that [hidden email] can be added without requesting
> permission.
>
> Is there more up-to-date information about PKS sync? What's the current
> best approach?

Note that the pks and SKS keyservers at kjsl.com have been down for
a while, but Javier will be copying that data back online for me
RealSoonNow(TM).  If you don't mind the bounces for a while longer,
please continue to keep [hidden email] in
./mailsync on all SKS servers.  Remember that this helps propagate
keys between the two keyserver networks as quickly as possible.

BTW, if anyone has a shell account (SSH public key) or two (esp. on
FreeBSD 6.x) to spare, I would like to be able to keep the keyanalyze
reports and perhaps a pks instance backed up online via rsync from/to
kjsl.com.  If you're already running SKS on port 11371, I wouldn't
necessarily need pks running on port 21371 or mailsync, esp. if I
can start using your SKS keydumps (every other Sunday) to feed
keyanalyze.  The archived reports (and my other web content) are
.75GB now and add ~10MB every other week w/o individual reports.
(ZFS would be helpful for snapshots/rollback of the BerkeleyDB
database files and keydumps, versioning keyanalyze data, and
calculating/propagating updates to other ZFS users w/o allowing
global rsync.)

If anyone wants to start putting the keyanalyze data into a
database (to track individual reports and/or MSDs) and/or
subversion or mercurial repo. (for the "permanent" files),
that would also be welcome.


On Mon, Oct 29, 2007 at 08:59:14AM -0400, Yaron Minsky wrote:

> I think the optimal thing would simply to have everyone in your peer list.
> I don't think you lose anything to that, but it's a pain to maintain.  I

This causes gossip attempts to be missed more frequently, unfortunately,
for example when trying to connect to keyserver.noreply.org to initiate.

--
Jason Harris           |  NIC:  JH329, PGP:  This _is_ PGP-signed, isn't it?
[hidden email] _|_ web:  http://keyserver.kjsl.com/~jharris/
          Got photons?   (TM), (C) 2004

_______________________________________________
Sks-devel mailing list
[hidden email]
http://lists.nongnu.org/mailman/listinfo/sks-devel

attachment0 (322 bytes) Download Attachment