Re: Benchmarking Methodology for SDN Controller Performance - Updated Draft Version

"Marius Georgescu" <[email protected]>
Newsgroups gmane.ietf.bmwg
Message-ID <[email protected]>
Dear Bhuvan et alia,

 

Please find attached my inline suggestions for the new draft. They were
marked with ###MG .

Thank you for taking care of so many comments.  

 

Best regards,

Marius 

 

 

From: bmwg [mailto:[email protected]] On Behalf Of Bhuvan (Veryx
Technologies)
Sent: Tuesday, July 21, 2015 8:23 PM
To: [email protected]
Cc: 'Anton Basil' <[email protected]>; Tassinari, Mark A
<[email protected]>; vishwas.manral <[email protected]>
Subject: [bmwg] Benchmarking Methodology for SDN Controller Performance -
Updated Draft Version

 

Dear BMWG Members,

 

We have updated the draft
(draft-bhuvan-bmwg-sdn-controller-benchmark-meth-00
<http://tools.ietf.org/html/draft-bhuvan-bmwg-sdn-controller-benchmark-meth-
00> ) about SDN Controller benchmarking addressing comments received in
IETF-92 meeting. Thank you very much for providing your valuable comments.
The latest draft can be found in
draft-bhuvan-bmwg-sdn-controller-benchmark-meth-01
<http://tools.ietf.org/html/draft-bhuvan-bmwg-sdn-controller-benchmark-meth-
01> 

 

Summary of Changes:

a.      Updated test setup diagram following the comment from Scott Bradner.

b.      Added recommendations for test topology, test iterations etc., to
use for benchmarking.

c.      Provided reference test topologies.

d.      Split Path Provisioning tests into two different tests - Proactive
and Reactive Path Provisioning tests.

e.      Provided more clarity on test procedure for some of the tests.

f.       Fixed IETF normative language usage.

 

We would love to hear any comments and queries on the same.

 

Thanks,

Authors

_______________________________________________
bmwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/bmwg
draft-bhuvan-bmwg-sdn-controller-benchmark-meth-01 MG.TXT (text/plain, 8.3 KB)
...


4.6 Test Repeatability

   To increase the confidence in measured result, it is recommended that
   this test SHOULD be performed atleast 10 times with same number of 
   nodes using same topology.

   ###MG : Reporting the repeatability probably will come up at some point. I would suggest reporting 
   the relative standard deviation. The calculation formula can be something like :
         sigma
rsd = ----------- x 100   
        avgScore

...
   
   
6.1.3 Asynchronous Message Processing Rate 
   
   Objective:
      To measure the maximum number of asynchronous messages (session
      aliveness check message, new flow arrival notification 
      message etc.) a controller can process within the test duration, 
      expressed in messages processed per second.

   Reference Test Setup:
      The test SHOULD use one of the test setups described in 
      section 3.1 or section 3.2 of this document.
   
   Prerequisite: 
      1. The controller MUST have completed the network topology
         discovery for the connected SDN nodes.  

   Procedure: 
      1. Generate asynchronous messages continuously at the maximum
         possible rate on the established connections from all the 
         connected SDN nodes in the forwarding plane test emulator
         for the Test Duration (Td).
      2. Record total number of responses received from the 
         controller (Nrx) as well as the number of messages sent(Ntx) to
         the controller within the test duration(Td) at the forwarding 
         plane test emulator interface (I1) .

   Measurement: 
                                                 Nrx 
      Asynchronous Message Processing Rate Tr1 = -----
                                                 Td
                                                   Tr1 + Tr2 + Tr3..Trn
      Average Asynchronous Message Processing Rate= --------------------
                                                  Total Test Iterations

      Loss Ratio = (Ntx-Nrx)/100.

###MG: There might be  mistake here. I guess the calculation formula should be something like this:

Loss ratio = (Ntx - Nrx/Ntx)*100 

...
   

6.1.8 Network Topology Change Detection Time

   Objective: 
      Measure the time taken by the controller to detect any changes
      in the network topology, expressed in milliseconds.
   
   Reference Test Setup:
      The test SHOULD use one of the test setups described in 
      section 3.1 or section 3.2 of this document.

   Prerequisite: 
      1. The controller MUST have discovered the network topology 
         information for the deployed network topology.  
      2. The periodic network discovery operation should be configured 
         to twice the Test duration (Td) value.





Bhuvan, et al.            Expires January 18, 2016             [Page 16]

Internet Draft    SDN Controller Benchmarking Methodology      July 2015


   Procedure: 
      1. Trigger a topology change event by bringing down an active 
         SDN node in the topology.
      2. Record the time when the first topology change notification
         is sent to the controller (Tcn) at the forwarding plane 
         test emulator interface (I1).  
      3. Stop the test when the controller sends the first topology 
         re-discovery message to the SDN node or the expiry of test 
         interval (Td).  
      4. Record the time when the first topology re-discovery message
         is received from the controller (Tcd) at the forwarding plane
         test emulator interface (I1)

   Measurement:
      Network Topology Change Detection Time Tr1 = Tcd-Tcn.

                                        Tr1 + Tr2 + Tr3 .. Trn
      Average Network Topology Change 
                      Detection Time = --------------------------- 
                                        Total Test Iterations

   Reporting Format: 
      The Network Topology Change Detection Time results MUST be 
      reported in the format of a table with a row for each iteration. 
      The last row of the table indicates the average Network Topology 
      Change Time.  
###MG: I've received this question for my draft as well(from Kostas), and and maybe it stands as a general question for the WG. 
Why  average (mean) and not median ? A short wording on why choose one over the other would probably do. I was thinking
to cite some of the published results and show that the probability distribution was normal. Hence, the mean and median
would be almost identical and the average has the advantage of including outliers, which I think are important for the overall
performance.

...

11.4 Performance Benchmarking Tests

11.4.1 Network Topology Discovery Time

   Procedure:

      SDN Nodes                     OpenFlow                    SDN    
                                   Controller               Application 
         |                            |                           |
         |                            |<Initialize controller     |
         |                            |app.,NB and SB interfaces> |
         |                            |                           |
         |<Deploy network with        |                           |
         | given no. of OF switches>  |                           |  
         |                            |                           |
         |    OFPT_HELLO Exchange     |                           |
         |<-------------------------->|                           |
         |                            |                           |
         |    PACKET_OUT with LLDP    |                           |
         |      to all switches       |                           |
    (Tm1)|<---------------------------|                           |
         |                            |                           |
         |         PACKET_IN with LLDP|                           |
         |          rcvd from switch-1|                           |
         |--------------------------->|                           |
         |                            |                           |
         |         PACKET_IN with LLDP|                           |
         |          rcvd from switch-2|                           |
         |--------------------------->|                           |
         |            .               |                           |
         |            .               |                           |
         |                            |                           |
         |         PACKET_IN with LLDP|                           |
         |          rcvd from switch-n|                           |
    (Tmn)|--------------------------->|                           |
         |                            |                           |
         |                            |      <Wait for the expiry |
         |                            |     of Test Duration (Td)>|
         |                            |                           |
         |                            |   Query the controller for|
         |                            |   discovered n/w topo.(Di)|
         |                            |<--------------------------|
         |                            |                           |
         |                            |   <Compare the discovered |
         |                            |    & offered n/w topology>|
         |                            |                           |




Bhuvan, et al.            Expires January 18, 2016             [Page 28]

Internet Draft    SDN Controller Benchmarking Methodology      July 2015

###MG : I have to say that these diagrams are helping a lot with understanding the benchmarking procedure. 
I haven't done this in my draft as well, but I think  adding some sample reports might be actually very useful.

   Legend:
      NB: Northbound
      SB: Southbound
      OF: OpenFlow
      Tm1: Time of reception of first LLDP message from controller
      Tmn: Time of last LLDP message sent to controller

   Discussion:
      The Network Topology Discovery Time can be obtained by calculating
      the time difference between the first PACKET_OUT with LLDP message
      received from the controller (Tm1) and the last PACKET_IN with
      LLDP message sent to the controller (Tmn) when the comparison is
      successful.

###MG : The name of the paragraph (Discussion) is a bit confusing, I think. I guess you want to detail the calculation
method. Discussion makes me think that you're still not sure that this is the way to calculate it. 
	  
...
lmpx.com only provides a reader for public news (NNTP) servers. It is not affiliated with the servers or forums shown here and is not responsible for the content of articles, which is written by their respective authors.