You do. ;-) If you have freeradius you have radiusclient. Ivan Kalik Kalik Informatika ISP Dana 4/10/2007, "Amr el-Saeed" <amr.elsaeed@tedata.net> piše:

Dear All,

Does any one have any tool to stress test the freeRadius ??

regards, Amr el-Saeed - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

Like most other RADIUS server load testers it merely tests how fast the client load test can run and doesn't really test the server load. The Evolynx tester is especially prone to this problem because you can't set the client time out. And 20 concurrent threads won't result in much of a load. And CPU load may not be much of a measure of a server's capacity. In a number of my own tests I've found that the shorter the time out the faster the server appears to run with only slightly more packets being dropped. What we're seeing is that a long time out can make it appear that all packets are being processed but the long time out also slows the client load test. If half of those threads are waiting a very short time for responses and half are waiting a long time (essentially waiting for the short responses to be processed by the server) you're now down to half the threads testing the server. As time goes by more and more threads are waiting for fewer quick thread responses and the test becomes meaningless. Being able to set the time out to something like one tenth of a second can show a significant increase in server packets processed with only a slight increase in dropped packets for simple authentication methods. A dramatic increase in packet loss usually shows a back end problem. The only real way to load test a server is to run many tests from many machines. For example, I saw the graph of over 100,000 customers using a RADIUS accounting server. The dropped packet count was sloping up gradually as the morning progressed and more people were using the system. Finally the back end server completely bogged down and then the dropped packets spiked enormously as customers started re-accounting causing a complete failure of the system from which it wouldn't recover until most of the customers gave up. That's the test you're looking for and that's the load you want :) Kostas Kalevras wrote:

O/H tnt@kalik.co.yu έγραψε:

...

You do. ;-)

If you have freeradius you have radiusclient.

Ivan Kalik Kalik Informatika ISP

Dana 4/10/2007, "Amr el-Saeed" <amr.elsaeed@tedata.net> piše:

...

Dear All,

Does any one have any tool to stress test the freeRadius ??

http://www.evolynx.com/radius/dl_loadtest.aspx

...

...

regards, Amr el-Saeed - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

Dear All, i know this is too late to say that :-) but , i used radclient to test my freeradius server i wanted to use the ( -p num Send 'num' packets from a file in parallel. ) thats my command ( radclient -f attr -p 100 localhost:1812 auth local ) this is the file attr ( User-Name = "amrali" User-Password = "123456" NAS-IP-Address = 255.255.255.255 NAS-Port = 34 ) but the request is done only once not 100 times !! any idea ? Amr Ali Alan DeKok wrote:

Michael Lecuyer wrote:

...

What we're seeing is that a long time out can make it appear that all packets are being processed but the long time out also slows the client load test. If half of those threads are waiting a very short time for responses and half are waiting a long time (essentially waiting for the short responses to be processed by the server) you're now down to half the threads testing the server. As time goes by more and more threads are waiting for fewer quick thread responses and the test becomes meaningless.

I'm looking into creating a complex load testing tool (not radclient), that will be single threaded, but which will be able to have many 1000's of outstanding requests. It will track a number of pieces of information, including offered load, accepted load, histograms of delay over time, etc.

It can be run on mltiple machines, and the results collated.

Ping me privately if this is interesting.

Alan DeKok. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

Dear All, i know this is too late to say that :-) but , i used radclient to test my freeradius server i wanted to use the ( -p num Send 'num' packets from a file in parallel. ) thats my command ( radclient -f attr -p 100 localhost:1812 auth local ) this is the file attr ( User-Name = "amrali" User-Password = "123456" NAS-IP-Address = 255.255.255.255 NAS-Port = 34 ) but the request is done only once not 100 times !! any idea ? my free radius version is freeradius-1.1.0-1 Amr Ali Alan DeKok wrote:

Michael Lecuyer wrote:

...

What we're seeing is that a long time out can make it appear that all packets are being processed but the long time out also slows the client load test. If half of those threads are waiting a very short time for responses and half are waiting a long time (essentially waiting for the short responses to be processed by the server) you're now down to half the threads testing the server. As time goes by more and more threads are waiting for fewer quick thread responses and the test becomes meaningless.

I'm looking into creating a complex load testing tool (not radclient), that will be single threaded, but which will be able to have many 1000's of outstanding requests. It will track a number of pieces of information, including offered load, accepted load, histograms of delay over time, etc.

It can be run on mltiple machines, and the results collated.

Ping me privately if this is interesting.

Alan DeKok. - List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

sorry but i was unsubscribed from the milling list, i don't know why ? so, i didn't get the answer !! can u just send it again if u not mind ?? thanks tnt@kalik.co.yu wrote:

...

any idea ?

You have asked this a couple of days ago. Try reading the answer.

Ivan Kalik Kalik Informatika ISP

- List info/subscribe/unsubscribe? See http://www.freeradius.org/list/users.html

Hi,

Hi All

i used " radclient -f attr -p 10 -c 1 localhost:1812 auth local "

but also got one entry in my log file !! i guess i must get 10 entries .

err, no. because ou only asked for it to do a single request ( -c 1 ) if you want it to do 10, then tell it to do 10. the -p 10 means to to all ten at once eg -c 1 -p 1 - one request, one request -c 10 -p 1 - ten requests, one at a time -c 10 -p 10 - ten requests, all at once -c 100 -p 10 - one hundred requests, ten at a time -c 100 -p 100 - one hundred requests...all at once! you can then increase the speed/intensity by using the -n flag.. but if you havent comprehanded the -c/-p link then leave -n alone alan

On Thu, 2007-10-04 at 14:39 +0200, Alan DeKok wrote:

wlan@mac.com wrote:

...

Unfortunately, from a back-end perspective, if having to support a variety of 'broken' commercial vendors, one doesn't have much choice (apart from not supporting these vendors). I guess we have to live with it as it would probably be difficult for these vendors to change their ways now...

Acct-Input-Octets has one meaning: the right one.

You don't have to interoperate with broken vendors. You tell users to throw the equipment away, and to buy working equipment.

Go get 'em! I couldn't resist any longer. RADIUS = Remote Authentication Dial In User Service A number of us on this list and historically on others have been developing radius server and related software. I can't remember any of the developers asking what "Input-Octets" or "Output-Octets" meant. --- snip --- 5.3. Acct-Input-Octets Description This attribute indicates how many octets have been received from the port over the course of this service being provided, and can only be present in Accounting-Request records where the Acct- Status-Type is set to Stop. --- snip --- This would be data coming into the NAS from the remote end point {customer}. --- snip --- 5.4. Acct-Output-Octets Description This attribute indicates how many octets have been sent to the port in the course of delivering this service, and can only be present in Accounting-Request records where the Acct-Status-Type is set to Stop. --- snip --- This would be data going from the NAS to the remote end point {customer}. Unless the reader or translator did not understand English or the parts of a Dial In service there should be no reason for it to be interpreted any other way. If a vendor is unable to comprehend a small concept like in and out, I would not trust them to properly resolve more tricky concepts, like refunds, so Alan suggestion stands. Throw out the improperly designed equipment if the vendor did not just make an error in their manual. If the error is in the manual and not the equipment, let them know, so they can publish an errata. Do a test, figure out which is correct the equipment or the manual.