[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Scalability of Netconf



On Mon, Jul 04, 2005 at 01:19:57PM -0400, Sharon Chisholm wrote:
 
> These would be interesting numbers to get. I've heard related (but
> opposite?) concerns around being able to get the most out of each
> netconf connection to ensure that more complex deployments don't max out
> on connections. This might eventually require us to continue the
> channelization discussions which was mainly deferred, but I think that
> can wait a bit. First step would be to determine if this is really an
> issue in practice.

I am not yet done with my experiments. So far, I have writen a script
which connects to ssh subsystems I called cat:

Subsystem       cat     /bin/cat

The script kicks off a standard ssh client to connect to a remote cat 
subsystem and then starts a loop where it periodically sends a message
and checks whether that is echoed back correctly. Nothing fancy. (Note
that this script is rather heavy weight since it requires a new ssh
process for each connection - a more serious implementation would
require some C code so that you can keep multiple connections in one
process.)

I am running this script on a 3GHz Pentium with 512 MB Ram (so really
nothing high end for a management system) running Linux 2.6.10. I 
keep 300 connections open and the funny thing is that this box does 
not at really get stressed by this. The load never went above 0.1% 
and was most of the time significantly below. The memory used excluding
buffers was ~235 MB.

Given that the script is really a crude resource hog, the box performed
much better than expected. The next step is probably to shoot for
1000 ssh connections with this crude script. I just have to setup
more cat subsystems to talk to...

/js

P.S. I admit that my Mac OS X notebook does not behave that friendly
     since I was simply running out of slots in the process table. ;-)

P.S. Attached is the ugly quick and dirty script - I am sorry for
     using Tcl and not Perl - I am sure Perl will make this faster. :-)

-- 
Juergen Schoenwaelder		    International University Bremen
<http://www.eecs.iu-bremen.de/>	    P.O. Box 750 561, 28725 Bremen, Germany

--
to unsubscribe send a message to netconf-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/netconf/>