network thruput

Tue May 31 20:52:44 PDT 2005

> - current upper management premise ...
> 	- keep windoze and backup machines on the same lan
> 	because it doesn't clutter the wire
> 	even thou 80% of the packets is udp and netbios :-) duuhhh...

Did you draw pretty pictures to dissuade them?  That usually works for
me...the more colors the better.

> - i claim, think, prefer, that, in order to maximize network performance,
>   i would:
> 	- put all windoze machines on its own WindowLAN ( 192.168.x.y )
> 
> 	- i would put all cluster machines on its own 192.168.clusterIP
> 
> 	- i would put all daily backup traffic on its own 192.18.backupIP

So, all machines have a minimum of two NICs?  That how we have our
network setup.  At a minimum each Unix server has a production NIC and a
backup NIC.  During the construction and planning phase of the
datacenter, they even planned for a monitoring/admin NIC for a total of
three NICs per host.  These third NICs/networks were never utilized. 
The more segregation the better, and quieter, the network will be.

With your above idea, why not have the backup server on both LANs? 
You'd be able to segregate even further.

> 	- we're stuck with the intel gigE nics and nortel switches
> 	and cat-5e but the stacked nortels is stacked as a single switch,
> 	which i think is a bad thing
> 
> 		- they're getting a measly/pultry 5-10MByte/sec at best
> 		on a gigE system .. which is why i'm "arguing" with
> 		those that have the power to say "bi-bi" to me :-)
> 
> 		- i think its a fun problem to solve.. but..
> 		and i already have scripts that show we can go
> 		4x or 5x faster with simple separate LANs
> 		but sometimes, i think the itty-bitty dual-2.x CPUs 
> 		just run out of memory that the private network not 
> 		always faster

I've never had a problem with any of our Dell's that have the
prod/backup NIC setup.

> 	- i'd have a top-level switch from each of the private LAN
> 	going into the "top level" switch to connect all the individual
> 	lans

This is also how we have our network setup.  Each switch has two
upstream switches/routers (depending at which point the switch/router
resides) for redundancy.  From there they go to a router (if the BFR/BFS
can't route).

eric