OK, well we're also seeing lots of these messages on the backend:

INFO: Duplicate request id 147 received from 128.248.155.31(41004): retransmit 
reply

What exactly do these signify? Could the UDP network queues be filling up on 
the backend servers? How much memory should be allocated to them?

Thanks for any help!

---
Roberto Ullfig - ACCC Research Programmer
[email protected]

________________________________________
From: Bjoern A. Zeeb [[email protected]]
Sent: Friday, September 05, 2014 4:16 AM
To: Ullfig, Roberto
Cc: [email protected]
Subject: Re: [RADIATOR] AuthRADIUS : Could not find a working host to forward 
messages

On 05 Sep 2014, at 06:39 , Ullfig, Roberto <[email protected]> wrote:

> Hello,
>
> with the semester just beginning we find that our Radius services (using UDP) 
> can't keep up. The servers themselves are pretty much idle. We're using 3 
> load balancers and AuthBy HASHBALANCE - connecting to 7 radius instances on 5 
> different servers (RHEL 6). We've been running like this for a few years. At 
> peak we're authenticating ~15 users a second but we're now seeing tons of 
> these messages (128231 on Thursday) on the load balancers and lots of users 
> can't connect:
>
> AuthRADIUS : Could not find a working host to forward messages
>
> We're having a hard time determining the bottleneck. Should we just add a 
> whole bunch of new instances on these idle servers?

Well a first good thing would be to determine why your load balancers and your 
“backend” Radius servers have trouble talking to each other or why the backends 
are not replying.  With a well-tune backend you should easily be able to get up 
to more than thousands of requests / s to handle on a single decent servers 
running Radiator these days.

Ideally start debugging from the back:  check what your 7/5 different servers 
and your backend think.  Turn debugging on for a few minutes on one of your 
servers, read the log files;  run tcpdump along at the same time.  If these 
servers and your backend perform check the same between the load balancers and 
radius backend servers.

My bet usually is on the backend having trouble for some reason, in which case 
more Radiators won’t do you any good.   What’s the backend your are using to 
lookup your users?


—
Bjoern A. Zeeb             "Come on. Learn, goddamn it.", WarGames, 1983

_______________________________________________
radiator mailing list
[email protected]
http://www.open.com.au/mailman/listinfo/radiator

Reply via email to