[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [K12OSN] Adding another server as backup.



On Wednesday 11 May 2005 08:49, R. Scott Belford wrote:
> Martin Woolley wrote:
> > As far as the load balancing goes, we did nothing special.  Each LTSP
> > server was setup the same.  Each is a DHCP server and each serves
> > addresses in the same network range.  The client PXE boots, grabs an i/p
> > address from which ever server responds first and is loaded with a
> > kernel. The second DHCP request (for the NFS) may be served by a
> > different server to that which supplied the kernel.
>
> I find this very interesting.  How evenly is the load balanced amongst
> the servers?  Have you done any statistical analysis?  Is natural order
> actually arising from this seemingly chaotic design?  It is very clever.

The statistical analysis we do is by counting how many login prompts are per 
server (we use KDE and you get the server name in the login box).  We can 
clearly see a randomish selection of our server names and the count is 
approximately equal. If there is no one on the network, and we reboot all of 
the clients, then the logins are all for one or two servers.  If we leave it 
like this, the two servers rapidly grind to a halt.   As more users logon 
(and network usage grows), if clients are rebooted, then the logins migrate 
amongst the servers.  We only reboot clients individially and then only if 
there is a problem with that client (eg a mouse is stolen or vandalised, 
which are two common problems).  Each night the servers are rebooted (a heavy 
handed method of ensuring that there are no zombie processes).  Next morning, 
you press the enter key on a client and it reconnects to the server that is 
was last conencted to.

How and why it all works is clearly related to the cleverness built into the 
DHCP specification and to the time slice that each process gets from the Unix 
kernel.  I vaguely remember reading that is there are two or more DHCP 
servers dishing out addresses in the same i/p range, they keep track of 
addresses given out by the other servers so that no duplicates occur.  What I 
think happens is that when a client does a DHCP request (a DHCP Discover is 
put on the wire), my guess is that the machine that happens to have dhcpd as 
the active process serves an i/p address (puts a DCHP Offer on the wire).  
The client responds with a DHCP Request and when the DHCP Acknowledge is 
broadcast, other DHCP servers take a note of the i/p address and delete it 
from the pool. 
-- 
Regards
Martin Woolley
ICT Support
Handsworth Grammar School
Isis Astarte Diana Hecate Demeter Kali Inanna


*************************************************************
This email and any files transmitted with it are confidential
and intended solely for the use of the individual or entity 
to whom they are addressed. If you have received this email 
in error please notify postmaster bgfl org

The views expressed within this email are those of the 
individual, and not necessarily those of the organisation
*************************************************************


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]