Re: Scaling to 1500 clients - any one done this...

[Chris Edillon]

On Wed, 2004-09-22 at 12:47, Chris Kacoroski wrote:

> I am scaling my install from 60 to 1500 clients.  At about 250 clients 
> I started seeing problems with server connections being dropped.  I 
> would like to hear from anyone else who has a install of equivalent size 
> to see how they set:
> 
> - how often the client connects (currently every 30 minutes)
> 
> - number of concurrent processes on the server (currently 100)
> 
> - Splaytime used (currently 15 minutes).
> 
  from a load-balancing perspective, do you really want
1500 clients connecting to a single server?  at a customer
site a few years ago i was running cfengine on over 1000
machines in various buildings, with a scheduled run every
half-hour and a splaytime of 25 minutes.  however, i had
the clients run a module which ran a simple heuristic to
determine which building it existed in and then set a
policyhost variable which pointed to the policyhost
located in the same building.  i never had any real
problems with that setup.  to mitigate the effect of
copying large amounts of data from the small set of
servers, i had a general purpose configuration which the
clients would pull down and run every half hour that only
handled local actions, and i imported a config file once
a night based on a time class in which the splay time was
set to four hours and all of the clients would pull over
software packages, etc. over the larger time period.
  unfortunately i'm consulting for a different customer
right now on a much smaller research project, so i don't
get to work on nifty large-scale configuration management
ideas using cfengine anymore.  i can describe more about
my old architecture off-list if anyone's interested.

chris