Weird issue - threads bottleneck in post-proxy section

[Alan DeKok]

On Mar 4, 2015, at 3:53 AM, RESTOUX, Loïc <loic.restoux at capgemini.com> wrote:
> Do you mean that you reach 20K pps with time-consuming processing in post-proxy ?

  I mean there’s 20K PPS with a carefully crafted configuration.  So FreeRADIUS *can* do 20Kpps proxying.

> It's a critical point, because it means that there is something bad in our overall environment, not in FreeRADIUS.

  Yes.  There is nothing in the server which is designed to destroy performance.  The server core is designed to run as fast as possible.

  But… if you use EAP (TLS), performance drops about 90%.  That’s due to the CPU overhead of calculating the RSA keys.  If you use a database, performance drops a similar amount.  Because databases have to write to disk, and disks are slow.

  I have *never* seen the server proxy 100 pps, even with complex rules in pre and post-proxy.  So there’s something weird about your local system.

  Alan DeKok.