Re: [bmwg] Feedback for Benchmarking Methodology for OpenFlow SDN Controller Performance

["Castelli, Brian" <[e-mail]>]

I really like Al¹s suggestion to generalize SDN controller benchmarking
because 1) the ONF effort is OpenFlow specific and 2) OpenFlow does not
define SDN. OpenFlow is one of several southbound protocols that can be
used between a controller function and network node functions. Regardless
of the maturity of current OpenFlow implementations, I believe that we can
define a family of benchmarks that have general applicability to SDN
environments because there are common functions/services that will be
provided independent of protocol. At the end of the day, protocols are
commodities. It¹s the service they provide that adds value.

The current draft contains many of the elements required to sufficiently
define a benchmarking methodology for a controller function in an SDN
environment. We should follow Al¹s suggestion, define the taxonomy used in
the draft, and consolidate wording to make the document easier to read and
apply.

What follows is an attempt to define a higher-level set of SDN controller
test cases. It builds on the work done in the draft. I will provide some
detail on the first test case in order to establish a cadence. Subsequent
test cases will not have the detail, but the ability to iterate over
variables and so on is implied.

The SDN Controller test taxonomy would include:

- Asynchronous Message Response Rate. A test case to measure AMRR would
cover a variety of conditions that are analogous to the following
OpenFlow-specific events:

  o Packet_in received
  o Flow expiration received
  o Link down received

In other words, AMRR would be used to measure the ability of an SDN
controller to respond to a variety of asynchronous (i.e. unexpected)
conditions. One methodology for measurement would be specified by the
benchmark. The tester would then apply that methodology to packet_ins,
flow expiration, link down, and whatever other AMR testing is appropriate.

The AMRR methodology should include iteration variables and goal seeking.
The tester ought to be able to use the methodology to determine, for
example, the maximum packet_in AMRR for a variety of conditions, such as
variations in the number of connected nodes, the complexity of the
packet_ins, and so on. Variables also include negative testing with
invalid packets, for example.

- Synchronous Message Response Rate. A test case for measuring SMRR would
cover a variety of conditions analogous to the following OpenFlow-specific
events:

  o Hello packet exchange
  o Echo request/response exchange

- Node Acquisition Rate. A test case for measuring NAR would be a special
case of SMRR, but it specifically measures the rate at which an SDN
controller can connect to nodes in a network.

- Node Acquisition Maximum. A test case for NAM would measure the maximum
number of network nodes that an SDN controller can connect to without
error.

- FailOver Response Time. There are at least two test cases for measuring
FORT. One would measure a control cluster¹s ability to survive a
controller failure. How long does it take the controller function to be
assumed by another controller element? Another would measure the
controller¹s ability to respond to node failures to do things like
re-route traffic. How long does it take for the controller to send out
commands that handle the error? This is a bit like measuring convergence
time, but care must be taken to keep the node variables out of the
measurement.

- FailBack Response Time. Test cases for FBRT would be used to measure the
response time when the controller comes back up or the node comes back
online. How long does it take the controller to send the commands that
restore service?



On 6/23/14, 10:53 AM, "Bhuvan (Veryx Technologies)"
<[e-mail]> wrote:

>Hi Al,
>
>Thanks for your comment. I do agree with you that there needs to be a
>common
>benchmarking mechanism for controller designs performing the same tasks.
>But the technology is still relatively immature and the approach has
>various
>dimensions including centralized control vs distributed control
>(controller
>less).
>Again the centralized control approach uses different programming methods
>(OpenFlow vs Non OpenFlow).
>
>Having said that, for better usability/understanding of the draft and to
>avoid misinterpretation, I personally feel it would be good to have
>separate
>benchmarking methodology for each approach (though the metrics remain
>same).
>Also we would be happy to continue the effort to extend the same metrics
>for
>other approaches too.
>
>Please let me know if I've missed something.
>
>Sarah/Brain, please also share your thoughts on this.
>
>Thanks,
>Bhuvan
>
>-----Original Message-----
>From: MORTON, ALFRED C (AL) [mailto:[e-mail]]
>Sent: Saturday, June 21, 2014 6:00 PM
>To: Castelli, Brian; Banks, Sarah; Bhuvan (Veryx Technologies);
>[e-mail]
>Cc: 'Anton Basil'; 'Vishwas Manral'
>Subject: RE: [bmwg] Feedback for Benchmarking Methodology for OpenFlow SDN
>Controller Performance
>
>Hi Brian, Sarah, Bhuvan, and all,
>
>a couple of quick answers and a comment:
>
>> -----Original Message-----
>> From: bmwg [mailto:[e-mail]] On Behalf Of Castelli, Brian
>...
>> My two main concerns are getting the terminology right and making sure
>> that the proposed set of tests will achieve our goals. I believe those
>> goals ought to include a common taxonomy, enabling apples-to-apples
>> comparisons, and minimizing the time/work required to execute the test
>> cases.
>
>Great, that's exactly what we are chartered to do in BMWG.
>http://datatracker.ietf.org/wg/bmwg/charter/
>
>
>>
>> I am willing to help improve the benchmark. I understand that it is an
>> early draft. The time is right to make sure that we are developing the
>> best specification that we can.
>>
>> How do we continue this process? Via email to this list? I am willing
>> to help.
>
>It's principally e-mail (just as you've been doing) and face-to-face
>meetings three times a year.  Sarah mentions that our next meeting is in
>her
>home town:
>http://www.ietf.org/meeting/90/index.html
>
>Make some text proposals for the draft, and we can discuss them on the
>list.
>
>Having said that, and recognizing that there appears to be a comparable
>activity in ONF that (I assume) is OpenFlow-specific, perhaps a way to add
>value to the industry is to approach the SDN controller problem more
>generically, such that different controller designs performing the same
>tasks could be benchmarked as black-boxes.  I realize this approach has
>substantial implications for the draft, but the benefit of wider
>applicability.
>
>food for thought,
>Al
>(as a participant)
>
>

Spirent Communications E-mail confidentiality.
-----------------------------------------------------------------------
This e-mail contains confidential and / or privileged information belonging to Spirent Communications plc, its affiliates and / or subsidiaries. If you are not the intended recipient, you are hereby notified that any disclosure, copying, distribution and / or the taking of any action based upon reliance on the contents of this transmission is strictly forbidden. If you have received this message in error please notify the sender by return e-mail and delete it from your system.

Spirent Communications plc
Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.
Tel No. [phone]
Fax No. [phone]

Registered in England Number 470893
Registered at Northwood Park, Gatwick Road, Crawley, West Sussex, RH10 9XN, United Kingdom.

Or if within the US,

Spirent Communications,
26750 Agoura Road, Calabasas, CA, 91302, USA.
Tel No. 1-818-676- 2300