[zeromq-dev] zeromq-dev Digest, Vol 34, Issue 124

Thijs Terlouw thijsterlouw at gmail.com
Wed Oct 27 16:56:36 CEST 2010


I seem to agree totally with Pieter on this. For me scalability means decentralize therefore auto-discovery and peer-to-peer communication with only a smallish number of peers. Scaling does not mean adding intermediate nodes automatically I believe.

I could imagine a scenario that Martin explains, but I think that's just one possible use case ( and less practical IMHO ). 

On Oct 27, 2010, at 22:27, Pieter Hintjens <ph at imatix.com> wrote:

> On Wed, Oct 27, 2010 at 12:25 PM, Martin Sustrik <sustrik at 250bpm.com> wrote:
> 
>> The problem I see is with large scale distributions. In a truly global
>> network each node can have potentially millions of peers. It's just not
>> viable to monitor presence of them all.
> 
> This is a false assumption, IMO.  Even in the largest networks a
> normal node never has more than a few hundred or thousand peers, and
> scaling is done by intermediation (devices) or simply by clustering.
> There are no current examples of networks afaik where nodes have even
> 1000s of peers, unless they fall into the "big fat server" class.
> 
> Typical examples of huge networks where nodes have few or several
> dozens of peers: Internet, SIP, bittorrent, human population.
> 
> You may find cases of nodes with 10K, 100K, 1M peers but these
> _exceptional_ and it is mistaken to equate this with scalability.
> It's almost the opposite.  Huge numbers of peers means massive
> centralization.  Distribution means nodes with reasonable numbers of
> peers and it's totally sane (and necessary) to monitor their presence.
> 
> Furthermore _every_ non-trival 0MQ application ends up doing presence
> monitoring, heartbeating or somesuch, in patterns where peers are not
> 100% passive.
> 
> -
> Pieter Hintjens
> iMatix - www.imatix.com



More information about the zeromq-dev mailing list