[p2p-hackers] scalability (was: p2p framework)

[Elias Sinderson]

Daniel Stutzbach wrote:

> I find it amusing when people claim a functioning 2-million peer 
> system doesn't scale. :-)

'Scalability' is relative, and the definition is often contingent upon 
the requirements of the system at hand. For example, an online banking 
application that could handle 2M active users would clearly be 
considered scalable enough for the application domain. However, when 
most people use the term these days, and especially with respect to P2P 
systems, they are using the term in reference to internet-scale 
scalability. To whit, while a 2M peer system may seem impressive at 
first blanche, a 20M peer system is better; a 200M peer system is _much_ 
better; a 2B peer system begins to approximate what we really want (at 
slightly less than half the size of the IPv4 address space).

The above represents a shift in scale of 3 orders of magnitude, 
something gnutella is clearly not capable of. In this sense, it is a 
tautology to say that gnutella doesn't scale. Sure, it may work well 
enough in a limited sense, but it is a fallacy to confuse that with true 
scalability. I mentioned the IPv4 address space, which is something like 
4.29B or so. This was thought to be more than enough addresses, yet now 
we find ourselves in the middle of shifting to IPv6, which has on the 
order of 3.4*10^38 addresses... Excessive, perhaps, but at least there's 
room to grow.  :-)

So, yes, Gnutella may scale well enough, but it clearly doesn't define 
(much loess aproach) scalability in any meaningful sense.


Regards,
Elias