[c-nsp] Network Monitor / Mgmnt App

Dale W. Carder dwcarder at doit.wisc.edu
Sat Jun 4 13:19:01 EDT 2005


On Jun 4, 2005, at 11:10 AM, Arturo Servin wrote:
>
>     We were using MRTG, then Cricket and now we are evaluating Torrus
> and RTG. Any comments about Torrus vs RTG?

Torrus (was RRFW) takes some of the configuration ideas from
Cricket, and uses a templated XML configuration system.  It
comes with a good range of discovery for well-known devices.
Since the config file is in XML, you have to recompile it to
make any changes.  In a network with a lot of monitoring churn,
it can take seemingly *forever* to do this operation.  Compared
to a usual config file, the XML system has a learning curve.

Torrus seems to have the same polling characteristics as running
multiple instances of MRTG, which is that it did not seem to
scale well in my testing.  If some polling operations get blocked,
expect the polling not to complete.  Torrus, like the rest of the
MRTG family, is not in my opinion for a big network unless
you throw a stupid amount of hardware at what is inherently
a software design problem.

RTG is a completely different beast than you're used to with
MRTG and friends.  There is a very high performance
multithreaded poller.  I did some tests and the PC sitting
on my desk at work could poll and record data for 15,000
64 bit instances across ~20 c6500 PE's in < 10 seconds.  The
data is stored into an SQL database.  RTG is lean.  It gets its
great performance characteristics by threading, randomizing
the polling order (so you don't pound on one box too much),
and by buffering disk writes via the SQL database.  Unlike
how most people use RRD, RTG's data store is non-lossy,
although if you wish, you can throw data away (average it).
Since there is no RRD file to update, you're not opening and
closing thousands of filehandles to do RRD's update & recalc
operation.

The last time I looked at RTG's graphical UI, it was a bit weak
compared to RRDtool, but I think that time series graphs are
mostly useless aside from placating management types anyway.
The SQL database lets you generate custom reports such as top-n,
histograms, or whatever else you can dream up, not to mention
the ability to search through your data!  There are some basic
examples that come with the distribution.  Configuration is
simple (or just as poor as MRTG), just script the config file
yourself and you're done.

Dale


----------------------------------
Dale W. Carder
Network Engineer
University of Wisconsin at Madison
http://net.doit.wisc.edu/~dwcarder



More information about the cisco-nsp mailing list