Collision Avoidance

[Rich]

Hi - We are having a terrible problem with performance
when we run TimeSlips. TimeSlips is running a Dell
PowerEdge server (running SBS2003) on a shared drive (T) -
locally the server's (E) drive. Our workstations (all
running XP Professional) have TimeSlips locally installed
(c:\program files\timeslips) and access the data file on
shared drive "S" also on the server's "E".
Our network is anchored by a Corvad router with
a "Collision" light - which glows RED when it detects a
collision.
When we work on non-TimeSlips applications or if we run
most of the TimeSlips functions the collision light does
not light. However, whenever we run the pre-bill
function (or some accounts receivable functions) the
light comes on and pretty much stays on. If we run the
pre-bill function strictly from the server we don't have
collisions (makes sense - no network activity) and the
response time is 3 or 4 seconds to do 15 clients. If we
run the pre-bill function from any client machine, the
collision light glows constantly and the 15 clients take
45 minutes.
There are a number of questions -
1. Any ideas about what is causing the collisions
and how to avoid them?
2. Is there a way for us to monitor the network
activity to see what is happening?
Any suggestions would be greatly appreciated.
Thanks
Rich

 

[Hans-Georg Michna]

When we work on non-TimeSlips applications or if we run
most of the TimeSlips functions the collision light does
not light. However, whenever we run the pre-bill
function (or some accounts receivable functions) the
light comes on and pretty much stays on. ...

1. Any ideas about what is causing the collisions
and how to avoid them?

Rich,

of course I can't be sure about anything without doing some
complex measurements, but my first guess is that there is
something wrong at a relatively low level of the network, like a
defective switch or trying to run network adapters set to full
duplex over a half duplex hub.

TimeSlips probably brings out the defect only because it creates
some (duplex?) network traffic.

2. Is there a way for us to monitor the network
activity to see what is happening?

There are measuring instruments for that, but they are expensive
and not easy to understand and use. I would first check the
switch, the one presumably built into your router. Is it a
switch at all? If not, run and get one. If you're happy with the
router, you could keep using it over just one Ethernet socket
and use a separate switch or several for the lines to the
computers.

Hans-Georg

 

[Rich]

Thanks Hans-Georg -

I appreciate your input. Forgive my ignorance, how do I
check to see if the network is running duplex or not.
While testing this out I moved the server and one desktop
to a separate Linksys router and ran TimeSips. The
problem still occurred.

I noticed that whenever I run the pre-build the network
traffic (as measured by task manager on the server)
spikes to 12% from about 0% (that is when no one else is
on the network but me) - the CPU is max 2%. Utilized. My
working theory is the problem has to do with the adapter
on the server (I am saying this because we know there is
no problem when we run strictly on the server and I have
changed the routers and cables.) The NIC card on the
server is an Intel Pro/1000 MT Network Connection. I
don't know what the parameters should be for the card.

Thanks for your help.
Rich

 

[Hans-Georg Michna]

I appreciate your input. Forgive my ignorance, how do I
check to see if the network is running duplex or not.
While testing this out I moved the server and one desktop
to a separate Linksys router and ran TimeSips. The
problem still occurred.

Rich,

the duplex problem only occurs when you have a half duplex hub,
not a switch. If you use a switch, then that can't be it.

I noticed that whenever I run the pre-build the network
traffic (as measured by task manager on the server)
spikes to 12% from about 0% (that is when no one else is
on the network but me) - the CPU is max 2%. Utilized. My
working theory is the problem has to do with the adapter
on the server (I am saying this because we know there is
no problem when we run strictly on the server and I have
changed the routers and cables.) The NIC card on the
server is an Intel Pro/1000 MT Network Connection. I
don't know what the parameters should be for the card.

Next think I'd do is change cables. It looks a bit like a
hardware problem to me. Then come the adapters. One of the
adapters, perhaps the one on the server, may be defective.

Changing the switch was a good idea. If the problem still
occurs, then it can't be the original switch.

Good luck---Hans-Georg

 

[Rich]

Hi Hans-Georg,

Thanks - I decided that I had to try to eliminate as many
variables as possible. I copied data from the server to a
second workstation. During the transfer, the network
utilization jumped to 6% as measured by the sever's
taskmgr(12% when TimeSlips is processing).
I then pointed the first workstation to the data on the
second workstation and tried the pre-bill again. This
time the pre-bill took less than two minutes for 18
clients - acceptable time - vs 20 mins on the server. The
second workstation shows 65% network utilization, but the
first workstation was pretty calm 1-2%.
So - I thought maybe the problem has something to do with
the physical data drive. I then moved the data to a
different drive on the server and shared that directory.
I ran the pre-bill and found no difference in performance
(20 minutes).
Since I am using a Dell workstation, Dell hub and dell
server - I will bring my findings to them and see what
they suggest.
Thanks for sticking with me.
Rich