Sure it upsets.
We have a bunch of average-populated 6500s,
using the default max age (which was, as far as I remember, 5) made
the switches very slow in responding to SNMP queries.
set them to 10, and, Gotcha! everything works very well.
ivan
I am just curios what kind of application/network requires this
aggressive monitoring.
Is it possible to share this information ?
Cheers
We use a several connections to the financial providers. This connections
are low bandwidth (up to 2 Mbps). This connections used by a number of front
end services from a nubmer of departments and we could not differentiate its
and configure QoS. But from time to time some one produce an extremely high
traffic spikes (less than 30 seconds) without congestion avoidance
mechanisms. Our task - is to find such applications and report to management
and developers a problem. Also if we'll be aware about it we could configure
QoS.
What kind of queuing are you using?
It sounds like configuring fair-queue on the interface (if your platform supports that, usually the ones with 2M interfaces do), it should help with the problem you're describing.
If you have CPU to spare, configure fair-queue everywhere you can where you don't have a "better" QoS-configuration in place. It really solves a lot of the problems people are seeing with FIFO and mixed traffic.
One place to start would be an open-source NetFlow collector/analyzer like nfdump/nfsen:
We are using cisco switches like as 3750, 6500 etc. So there is no
fairqueue.
How are the 2M connections connected to these ethernet switches?
Makes me wonder if it wouldn't be a better opex solution to actually put a CPU platform between the L3 switches and these 2M connections so you can do hierarchical shaping and prioritize the traffic you really need (and do fair-queue at the same time). An 1841 would do the job for instance.
I work on this way too. There ais no problem with netflow-sensor. But I can
not find good inexpensive collector for Windows which can collect data and
do graphic report.
My experience shows that Sergey is representative of shops in the financial sector. We have a number of clients who use OpenNMS to collect interface traffic data every two seconds for their links to trading systems.
-jeff
Open-source = free.
And you should be using *NIX, anyways. Using it for a simple project like this is a good learning experience.
;>
Yes you are correct about financial sector.
Is it possible to view flows (at least srs and dst addresses) in the NMS or
only interface utilization?
There is no problem with *NIX from the point of view qualification. But
corporate politic use only Windows servers and no any other OS in the
production.
They obviously use IOS or JunOS or what-have-you on their routers and other networking gear - classify this server as a piece of infrastructure equipment, and you're golden.
;>
In OpenNMS? No flow or conversation support built in as of today. Some have successfully integrated with cflowd, jflow, or other similar packages; I'm not familiar with the details of those integrations.
-jeff
> Our task - is to find such applications and report to management and developers a problem. Also if we'll be aware about it we could configure
> QoS.One place to start would be an open-source NetFlow collector/analyzer like nfdump/nfsen:
I use these tools with great success and can recommend them for a quick,
easy setup and trouble free operation. Combined with a few Linux based
internal gateways using fprobe-ulog (http://fprobe.sourceforge.net/) and
you can get a good picture of what's happening on your network.
This page may provide some guidance:
-----------------------------------------------------------------------
Roland Dobbins <rdobbins@arbor.net> // <http://www.arbornetworks.com>Sell your computer and buy a guitar.
LaDerrick
I wonder wether your are allowed to use cygwin on your windows machines;
that way you'd
might find http://qosient.com/argus/ helpfull;
cheers,
teemu