perfSONAR User Q&A and Other Discussion

["Garnizov, Ivan (RRZE)" <>]

Hi Devine,

 

“I didn’t see any packet loss since l was using iperf TCP. I will try to do the test using UDP to see the difference.”

It seems to me though you are not using perfSONAR as the tool for the measurements, but you have set yourself for iperf only. Since if you were using perfSONAR the results of the tests would have been recorded in the database and you wouldn’t have to run separate CLI test to demonstrate us the problem. It would be enough to share the graph or print it in PDF.

The biggest advantage that you would have with perfSONAR automatically is with iperf3, which would allow you to see retransmits (which definitely affect the performance in the net). Also you can easily set and record the OWD behavior.

If you want to stick with CLI, I suggest you switch to iperf3.

 

Best regards,

Ivan

 

 

 

From: Tshabalala, Devine (Mr) (s210051701) [mailto:]
Sent: Montag, 22. Juni 2015 15:54
To: Garnizov, Ivan (RRZE);
Cc: Szymon Trocha; Roderick Mooi; Shukri Wiener; Kevin Draai; Kudyachete Gratitude (Mr) (Summerstrand North Campus)
Subject: Re: Clarity on Iperf Results Collected

 

Good day

 

Thank you for the feed back. The environment is totally isolated and there is no other traffic passing. The figure below shows my laboratory set-up.

Do you use sampling for NetFlow?: I am using standard netflow v9 which export IP flow every 5minutes interval.

 

About the iperf time frame, I did run iperf for about 2hrs and recording results after 300 seconds of which the results looked the same with that on the screen shot. The iperf screen shot shown is for problem clarification purpose; otherwise the long test gave me the same results.

 

Did you see any packet loss in the traffic? I didn’t see any packet loss since l was using iperf TCP. I will try to do the test using UDP to see the difference.

Seems to me Netflow stats also account for the TCP overhead. So I would guess you can achieve better = similar results with UDP tests. I will try do another test utilizing UDP  for both ostinato packet generator and iperf. 

 

Thank you

 

 

Kind regards

DevineTshabalala

Masters Student

Department School of ICT : l.T. Communication Networks

(North Campus)

Tel: 078 491 6130

 

A successful man is one who can lay a firm foundation with the bricks others have thrown at him and

Action is the foundational key to all success.

 

 

---

From: Garnizov, Ivan (RRZE) <>
Sent: Monday, June 22, 2015 2:48 PM
To: Tshabalala, Devine (Mr) (s210051701);
Cc: Szymon Trocha; Roderick Mooi; Shukri Wiener; Kevin Draai; Kudyachete Gratitude (Mr) (Summerstrand North Campus)
Subject: RE: Clarity on Iperf Results Collected

 

Hi Devine,

 

Please clarify more about your environment and the measurements:

We can only assume here about your link speed and devices. Should we also assume that the environment is totally isolated and there is no other traffic passing?

Are you using switches or direct connections to the router? Could there be a bottleneck at the switch level.

 

The first thing that should be noted that you are comparing different timeframes. Iperf uses seconds and the nfdump aggregates for 5 minutes.

You do not share here how you are achieving 97.5Mb/s on 5 minute intervals, since your measurement with iperf lasted only for 30s.

 

Please also share the results from the iperf tool, where it failed to measure correctly the available bandwidth.

 

“Furthermore, according to my theoretical understanding, iperf fills the gap of unutilized bandwidth.” This is correct in case of TCP measurements, but it seems to me Netflow stats also account for the TCP overhead. So I would guess you can achieve better = similar results with UDP tests.

 

Best regards,

Ivan

 

 

 

 

From: [] On Behalf Of Tshabalala, Devine (Mr) (s210051701)
Sent: Montag, 22. Juni 2015 02:04
To:
Cc: Szymon Trocha; Roderick Mooi; Shukri Wiener; Kevin Draai; Kudyachete Gratitude (Mr) (Summerstrand North Campus)
Subject: [perfsonar-user] Clarity on Iperf Results Collected

 

Hi

 

I am trying to do some experiments using iperf and netflow tools. Therefore l tested my point to point link to create a network baseline using iperf tool. The results on figure below were obtained. In addition , l configured nfdump as my netflow collector and nfsen for graphical presentation. My question is; why is there difference in results presented by iperf tool and nfsen though it is the same link. My expectation was to see the results reflected on iperf command line to be similar to that on nfsen.

 

 

Furthermore, according to my theoretical understanding, iperf fills the gap of unutilized bandwidth. So, l generated some traffic from one end device to other transferring packets at a rate of roughly 47.5Mbps average. After some time, l run iperf tool at the same time to see its effect on the normal network transfer rate. l was expecting iperf results to be roughly be +- 50Mbps ( i.e. filling the remaining gap). Unfortunate, iperf results were lower than my expectations as indicated on the below figure. What could be the reason for such difference, or am l interpreting my results wrongly?

 

If it is meant to be that way, therefore my interpretation will be 47.5Mbps + 15.3 Mbps= 62.8 Mbps, hence leaving me with a question of; 97.5Mbps - 62.8Mps = 34.7Mbps (where did it disappear to). Please help.

 

Thank you

 

Kind regards

DevineTshabalala

Masters Student

Department School of ICT : l.T. Communication Networks

(North Campus)

Tel: 078 491 6130

 

A successful man is one who can lay a firm foundation with the bricks others have thrown at him and

Action is the foundational key to all success.