How to improve Radius Performance

Alan DeKok aland at deployingradius.com
Thu Dec 24 09:28:03 CET 2009


Tejesh Vijayakumar wrote:
> "time /usr/local/bin/radclient -q -s -f radius.test localhost auth testing123"
...
> real    1m55.536s
> user    0m28.046s
> sys     1m12.682s
> 
> Also i have commented 'unix' in raddb/sites-available/default file.
> Have attached the log below. Pls Note: The log is not complete for 10000 users. Just pasted log for few users getting authenticated.

  Well... here's a run on my machine (2.4G Core duo), with the server
running in debug mode, *and* the client tool (radperf) running in debug
mode.  The client is sending auth requests, followed by an accounting
start, and then an accounting stop.

  The "users" file sets "Auth-Type = Accept".  Other than that, it's a
stock configuration.

---
	     Total sent        :  30000
	     Total retransmits :  0
	     Total succeeded   :  30000
	     Total failed      :  0
	     Total no reply    :  0
	     Total time (s)    :  16.587
	     Packets/s         :  1809
	     Response times:
		< 10 usec  : 0
		< 100 usec : 0
		< msec     : 27505
		< 10 msec  : 2382
		< 0.1s     : 111
		< s        : 2
		< 10s      : 0
		< 100s     : 0
---

  That's nearly 2K packets/s, *with* full debugging on both the client
and server.  When I use "radclient", with only authentication packets
and full debugging on both, I get ~1300 packets/s.

  Since you're getting only ~80 packets/s, there's something seriously
wrong with your system.  All I know is that I can't see it here, or on
any of the machines I have access to (Mac, Solaris, Linux, BSD).

  I suspect it's not a RADIUS issue.

  I think that for 2.2.0, this performance issue will need to be
addressed.  We'll need to have some performance measurements inside of
the FreeRADIUS server itself, so that people can see where all of the
CPU time is going.  That will let people know what to blame.

  Alan DeKok.



More information about the Freeradius-Devel mailing list