[ILUG] [OT] Network Consultant

[Chris Higgins <chris.higgins>]

> Wow - I'm impressed - I thought I was being very vague :-) - That is exactly 

Seen it far too many times before - can spot it a mile away :)

> the problem - software broken, ppl who wrote it claim it is a network 
> problem....

But of course :)
> 
> > What about wandering over to the computer center and asking them - there
> > were a few damn good techies there when I was in UCC (but that was a while
> > ago :) Unless of course they are the people who told you it was your
> > network that was broken :)
> >
> 
> No I can't do that - I need an independant consultant that I can honestly say 
> I have no professional relationship with (I work in UCC).
> 
> > If you give us more details of what constitutes 'working correctly' and
> > what the network is comprised of we can be of more help - if it's a
> > cisco based network - then you'll want cisco certified people (CCNP /
> > CCIE)... if it's based on horse drawn carriages then you'll need a
> > farrier...  :)
> 
> <grin>
> Well it is only a small network of 5 computers with 100-base-T Ethernet 
> hub/nics - I kinda need ppl with general network experience - i.e. that can 
> test the physical links, the cards and the hub and certify they are working 
> correctly.

Erm - what OS ? 

Do you have a Unix machine anywhere there ? 

Is there any intelligence in the hubs ? (or are they truly dumb as a ditch ?)

Simplest way to identify if there is an intermittent problem is :
a/ Clock sync *everything* so all logs can be easily matched and cross 
referenced
   There should be at least one NTP server left in cs.ucc.ie, if not 
   a couple of others around the college

b/ Clear all counters on any interfaces / switch ports

c/ Setup one machine (possibly Linux/solaris) to run netsaint and monitor
   availability of all machines

d/ Setup same linux/solaris box to run mtr (http://www.bitwizard.nl/mtr/) 
   against every machine

e/ Sit back an wait for the software to fall over

Match your logs and look for anything odd on the network at the time
of the software dying. (ie: did you get dropped packets to any machine)

On a local lan you should have 0% packet loss  (especially for 5 machines on
a 100Mb ethernet network)

Check all of the interfaces that you have error / stats counters on and see
if there are any faults reports

- 

If this problem really exists, then having continual automatic monitoring is 
the
only way to find it - bringing in a consultant may not spot the problem... but 
if
you have a real problem that you can see (one of the machines is getting 10%
packet loss over the period of a day) then you have something firm to get the
consultant to find.


If you have this running for a whole week, and you can show the software 
company
that you have 100% availability of the network for all machines for a whole
week (168 hours) and that their software failed 10 times during this period - 
you put yourself in a better position to argue the case.



Anyway - what kind of failure are you seeing on the software ? Any well written
software (in the internet age) should handle packet loss - especially if they
are using TCP.

> 
> The software ppl are suggesting that our network cards are dropping 1k "now 
> and then" causing their software to fail - me getting strong unpleasant smell 
> :-)

Can they demonstrate this packet loss ?

What kind of cards are they ? 

> 
> thanks,
> tom.
> -- 
> regards,
> 
> Thomas B. Quillinan
> Centre for Unified Computing (http://www.cuc.ucc.ie)
> Go to http://www.csn.ul.ie/cgi-bin/finger.cgi?tom for my PGP Key
> 
> -- 
> Irish Linux Users' Group: ilug at linux.ie
> http://www.linux.ie/mailman/listinfo/ilug for (un)subscription information.
> List maintainer: listmaster at linux.ie

-- 
** Chris Higgins                         e: chris.higgins at horizon.ie **
** Technical Business Development        tel: +353-1-6204916            **
** Horizon Technology Group              fax: +353-1-6204949            **