huge news system architecture

[Marek Iwanejko]

Hello all,

I don't know which architecture will be better for heavy-loaded news
system.
Let's say that news system handles x*1000 concurrent NNRP users,
has text & binary groups and 2 TB storage.

Let's talk about avdantages and disadvantages of ideas:

1) 2-3 Heavy machines (let's say 4 CPU, 12GB RAM) as reader server(s)
   with disk arrays. One server is master, rest are slave.
   It's expensive, but full redundant, nice to maitenace.

2) Backend: 1-2 machines (2 CPU, 4GB RAM) with disk arrays
   and 4-8 small machines as caching frontends.
   It can be much more cheap, but I'm afraid about performance
   problems, scalability, active synchonizing, etc...
   Backend should deliver content to frontends via NNTP or NFS?
   What would be more powerfull?


I saw the presentation about designing big news system based on
Diablo and there was 2-nd solution. Here:
	- http://www.nntp.sol.net/sane2000/
	- http://www.xs4all.nl/~scorpio/news.jpg (related diagram)

...but it's rather theoretical presentation :(

I am die-hard fan of INN on reader server...
...so anybody could make a comment about these ideas, write about
potential problems or give URLs that I could read about similiar solutions
on INN?

Regards,
 Marek

-- 
_________________________________________________________________
Marek Iwanejko * M.Iwanejko at elka.pw.edu.pl *** in arp we trust **