[K12OSN] Network Error Troubles Again ...

[Terrell Prude', Jr.]

Jim Christiansen wrote:

> I know I'm shooting in the dark but my school needs some new ideas 
> again :-)
>
> My son is doing a Science Fair Project on clustered computers and has 
> everything functioning just fine and looks like he may make it to the 
> nationals with his angle using open mosix.
>
> He wanted to run a series of seti jobs on one of my labs that is 
> k12ltsp based but using a bunch of knoppix-cluster cds to come up with 
> some really impressive numbers (and for the fun of it).
>
> What he discovered was that of 14 network drops, only three 
> consistently returned errors:0 dropped:0.  All of the others after 
> running for only a minute or so were ifconfiging errors:25 dropped:150 
> or worse.  This is on the exact same machines that work steadily 
> through every school day as k12ltsp clients w-i-t-h one or two random 
> freeze-ups each day.
>
> Our technician has been in the school recently and has added a couple 
> new long runs to switches 300 feet away from the main gear in our 
> network.  I don't think that a network tester was used to check the 
> new runs...
>
> Is it possible that these errors be caused by new but faulty distant 
> network runs? Or, is this a problem that is more likely local to a bad 
> switch the library boxes are hooked to?
>
> Please feel free to bring out the crystal balls for this one!
>
> Jim 


Yup, I've seen this many, many times when things haven't been done just 
right.  Yes, the spec lets you go 328 ft (100m) with UTP, but I wouldn't 
do it in practice.  If there's even *one* fluorescent bulb nearby on 
such a long run, you're likely to have problems.  Why they didn't do a 
fiber run in that situation is beyond me; I always spec fiber when I'm 
getting to those distances, and definitely so if there are any 
RF-producing devices (say, fluorescent bulbs) along the pathway.

Running Cat 5e 300 feet to a drop meant for a switch is, to me, lunacy 
anyway, just from a bandwidth-usage point of view.  That should've been 
a fiber run with Gig-E on both ends.  No, it's not as expensive as it 
used to be; we do this all the time.

Perhaps the reason those random freeze-ups on the K12LTSP clients is 
occurring is because of the errors on those drops.  If you get enough 
latency with X11 (and it doesn't take that much), you hose your 
connection.  It's like if you were to temporarily disconnect and really 
quickly reconnect your client's Ethernet connection; you'd better hope 
you saved your data.  One or two random freeze-ups should not be 
happening in a K12LTSP setup; it sure doesn't on mine.  Have 'em check, 
and re-punch, if necessary, every single one of those drops.

HTH,

--TP