[bmwg] Objections to Terminology draft: OSPF convergence benchmarking

["Manral, Vishwas" <[e-mail]>]

Hi Scott,
 
Thanks a lot for the comments.
 
I will try to answer the comments based on each draft involved, please let
me know if you agree. Firstly we have had similar methodology used by
vendors(including us) to measure benchmarking our OSPF implementation.
Similar methodology has also been used in the and the results were presented
in infocomm measurement workshop. Check " Experience in
<http://www.cse.ucsc.edu/~aman/imw02-talk.ppt> Black-box OSPF Measurement" -
Albert Greenberg and Aman Shaikh.
 
Let me start with the terminology draft.
 
1) We have very clearly stated in the draft that the draft defines
terminology used in the Methodology and the applicability drafts.
 
2) We have defined only those terms which have rellevence to convergence.
The value add in this case is that we have added a discussion and relvence
to the methodology draft. 
 
For example Router Dead Interval effect how fast an OSPF implementation at
the other end knows that a neighbor has gone down, which clearly has an
effect on convergence. Similarly Hello Interval has been defined for it
effects how fast a router knows its neighbor comes up. Besides having an FE
interface as Point-to-point, will have different convergence values from
that if it is defined as broadcast.
 
3) Incremental SPF is defined in RFC2328, section 16.5 and 16.6(though only
for Type3 and Type 5 LSA's). Though a vendor can put tweaks to the
algorithm, I am not sure what your objection to defining the term is. We
have not defined how to do it, we have just defined the term Incremental
SPF.
 
4) Scott, regarding the term convergence we have very clearly stated that we
are testing protocol convergence, we are not testing dataplane
convergence(for which various different issues come up). We however have
been working on a draft for dataplane convergence for all protocols for the
last few months(not just IGP's), and we will forward that to the list once
we get through with the testing being done.
 
I will reply to the other objections of yours on the other drafts, and
probably comment what your draft does not address.
 
Thanks,
Vishwas

-----Original Message-----
From: Scott Poretsky [mailto:[e-mail]]
Sent: Friday, February 14, 2003 11:28 PM
To: [e-mail]
Subject: Re: Fwd: Re: [bmwg] WG Last Call: OSPF convergence benchmarking


Folks,

While I think that IGP Convergence testing is very worthwhile work for the
BMWG to undertake, these drafts do not address IGP Convergence.  These
proposed drafts focus on LSA propagation benchmarking.  Due to the extreme
interest in industry for IGP convergence measurements, I was working on IGP
Convergence terminology and benchmarking methodology drafts when these
proposed drafts were submitted.  I have provided the IGP Convergence drafts
to Kevin and they should soon be posted.  My comments to the proposed LSA
Propagation benchmarking drafts are below.  

Scott
----------------------------------------------------------------------------
-
Terminology
Internal Measurements and External Measurements are already defined as White
Box and Black Box.

Point-to-Point, Broadcast, SPF, Hello Interval, and Router Dead Interval are
already defined in IETF RFCs

Incremental-SPF is a vendor-specific feature.

Convergence is stated as " A network is termed to be converged when all of
the devices within the network have a loop free path to each possible
destination. "   The definition is then modified for  a single DUT as follow
"Since we are not testing network convergence, but performance for a
particular device within a network, however, this definition needs to be
narrowed somewhat to fit within a single device view.  In this case,
convergence will mean the point in time when the DUT has performed all
actions needed to react to the change in topology represented by the test
condition ".   This must include updating the FIB and hardware as externally
verified with data traffic.  This draft contradicts its own definition by
verifying convergence through measurement of only LSA propagation.  In fact,
updating the FIB is explicitly excluded from measurement.

Methodology
Section 5. Overview and Scope
Control Plane measurements are not convergence measurements.  Convergence is
benchmarked by time to reroute traffic.

Section 6. Test Conditions
Bullet 2  It cannot be assumed that there is zero delay to execute the SPF
or  LSA processing.  Reference [5] actually _recommends_ router vendors
implement zero delay to achieve low convergence.  The benchmarking
methodology should allow convergence measurements to be made regardless of
implementation.

Bullet 3  Data traffic is actually required to make a deterministic
measurement of route convergence.
Unlisted: Number of IGP routes is a critical parameter for convergence time.

Sections 7. through 9.
The biggest problem with the methodology is that not all components for
convergence time are measured.  Focus is on LSA Processing, which is only 1
of 4 major components.

Route change due to cost or next-hop change is not considered.  This is the
most common reason for route convergence.

Well -defined Convergence methodology can be applied to ISIS or OSPF.  The
Link-State IGP does not matter.  In fact the methodology should be able for
use to benchmark one IGP against the other given the same DUT.

All procedures meet one or more of the following flaws:
White box measurements are part of the procedure.
Time to install route (update FIB), update hardware, and reroute traffic are
not considered.
Test case is Implementation-specific.
Test case has nothing to do with convergence - more OSPF performance test. 

Applicability
The applicability document explains that convergence testing is needed using
1. White box benchmarks
2. DUT intrusive measurements
3. Knowledge of DUT implementation
4. Absence of traffic results in impractical benchmarks.  Convergence time
is benchmarked by traffic loss.
1 through 3 violate basic test methodology requirements.  4 makes invalid
convergence tests.
This work is less about convergence benchmarking and more about LSA
propagation benchmarking.




At 08:42 PM 2/7/03 -0500, you wrote:



BMWG'ers:

A WG Last Call period for the Internet-Drafts regarding
OSPF convergence benchmarking terminology, methodology, and
benchmark applicability:

   <draft-ietf-bmwg-ospfconv-term-02.txt>,
   <draft-ietf-bmwg-ospfconv-intraarea-03.txt>,
   <draft-ietf-bmwg-ospfconv-applicability-01.txt>

will be open from 7 February 2003 until 28 February 2003.

Please weigh in on whether or not you feel each individual draft
should not be given to the Area Directors for consideration in progressing
the draft to an Informational RFC.  Send your comments
to this list or [e-mail].


URLs for the Internet-Drafts are:

http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-term-02.txt
<http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-term-02.txt> 
http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-intraarea-03.tx
t
<http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-intraarea-03.t
xt>  
http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-applicability-0
1.txt
<http://www.ietf.org/internet-drafts/draft-ietf-bmwg-ospfconv-applicability-
01.txt>  

-Kevin


_______________________________________________
bmwg mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/bmwg
<https://www1.ietf.org/mailman/listinfo/bmwg>