[bmwg] AD Review of draft-ietf-bmwg-evpntest

[Warren Kumari <[e-mail]>]

Hi all,

Unfortunately, I do not think that this document is ready to be
progressed -- there are a large number of areas where the document is
*very* unclear, and / or incomplete. I usually break my reviews up
into "substantive" and "Nits / Editorial"; but in this case I didn't,
largely because many of the editorial-type issues make the text
unclear; there are also a sufficient quantity that I do not feel
comfortable asking all of the IETF to review, nor would I feel
comfortable asking the directorates / IESG to review.

I've annotated my comments with "WK: ", but many of the issues occur
multiple times -- for example, there are many instances of "SA", which
from context is "Source Address" but this could also be "SA" in the
"Single-Active" sense.

W
-------------------


Abstract

   This document defines methodologies for benchmarking EVPN and PBB-
   EVPN performance.  EVPN is defined in RFC 7432, and is being deployed
   in Service Provider networks.  Specifically this document defines the

WK (nit) : Specifically, this (missing comma)

   methodologies for benchmarking EVPN/PBB-EVPN convergence, data plane
   performance, and control plane performance.



1.  Introduction

   EVPN is defined in RFC 7432, and describes BGP MPLS- based Ethernet
   VPNs (EVPN).  PBB-EVPN is defined in RFC 7623, discusses how Ethernet
   Provider backbone Bridging can be combined with EVPNs to provide a
   new/combined solution.  This draft defines methodologies that can be
   used to benchmark both RFC 7432 and RFC 7623 solutions.  Further,
   this draft provides methodologies for benchmarking the performance of
   EVPN data and control planes, MAC learning, MAC flushing, MAC ageing,

WK: s/ageing/aging  (RFC index has 258 instances of 'aging' and only 4 'ageing')

   convergence, high availability, and scale.

1.1.  Requirements Language

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this
   document are to be interpreted as described in RFC 2119 [RFC2119].

WK: Replace RFC2119 with RFC8174 language.



WK: Please fix formatting, and sort alphabetically. Also, this isn't
"Terminology", it is mainly Abbreviations.

1.2.  Terminologies

   MHPE Multi homed Provide Edge router.

   RR Route Reflector.

   P Provider Router.

   CE Customer Router/Devices/Switch.

   MHPE2 Multi homed Provider Edge router 2.

   MHPE1 Multi homed Provider Edge router 1.

   SHPE3 Single homed Provider Edge Router 3.

   AA EVPN Terminologies AA All-Active.

   SA EVPN Terminologies SA Single-Active.

   RT Router Tester.

   Sub Interface Each physical Interfaces is subdivided in to Logical
   units.
WK: s/in to/into/

   EVI EVPN Instances which will be running on sub interface or physical
   port of the provider Edge routers.

   DF Designated Forwarder.

   ESI Ethernet Segment Identifier.

2.  Test Topology

   EVPN/PBB-EVPN Services running on SHPE3, MHPE1 and MHPE2 in Single
   Active Mode:


         | [Traffic Generator ] Router Tester traffic sender/receiver
of layer 2 traffic with multiple vlan.

WK: VLANs. Also, this doesn't really parse - is 'Router Tester'
supposed to be in the brackets?

+----------+
|          |
|  SHPE3   |
|          |
+----------+
    |
    |Core link
+----------+
|          |
|  RR      |
|          | Route Reflector/Core router

WK: Core router or Provider router? (Core router isn't defined in the
terminology section)

+----------+-------------|
   |                     |
   |     Core links      |
+----------+       +-----------+
|          |       |    MHPE2  |
|   DUT    |       |           |
|  MHPE1   |       |           |
+----------+       +-----------+
     |    PE-CE link    |
+----------+------------
|          |
|  CE      |
|  layer2  |
|bridge    |
+----------+------------ [Traffic Generator](Router Tester
sender/reciever of layer 2 traffic with multiple vlan)

WK: As above. Aslo, receiver is a typo


WK: The formatting of this table is very poor, esp the header.



+-----------------+---------------------+---------------------+---------------------+----------------------+-----------------------+
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
| Mode            |                     |                     |
             |Receiver              |                       |
|                 |  Test               |Traffic Direction    |Sender
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |                      |                       |
+----------------------------------------------------------------------------------------------------------------------------------+
|                 |                     |                     |
             |                      |                       |
|                 |                     |                     |
             |      SHPE3           |                       |
|Single Active    |  Local Mac          |                     |CE
             |                      |Layer 2 traffic        |
|                 | Learning            | Uni                 |
             |                      |                       |
|                 |                     |                     |
             |                      | multiple MAC          |
|                 |                     |                     |
             |                      |                       |
+-----------------------------------------------------------------------------------------------------------------------------------+
|                 |                     |                     |
             |                      |                       |
|Single Active    | Remote MAC          |                     |
             |         CE           |Layer 2 traffic        |
|                 | Learning            | uni                 | SHPE3
             |                      |                       |
|                 |                     |                     |
             |                      |multiple MAC           |
|                 |                     |                     |
             |                      |                      ++
+----------------------------------------------------------------------------------------------------------------------------------+
|                 |                     |                     |
             |                      |                       |
|Single Active    | Scale Convergence   |   Bi                |
             |  CE/SHPE3            |                       |
|                 |                     |                     |
CE/SHPE3          |                      |Layer 2 traffic        |
|                 | Local& Remote       |                     |
             |                      |multiple mac& vlans    |
|                 | Learning            |                     |
             |                      |                       |
+-----------------+---------------------+---------------------+--------------------------------------------+-----------------------+

             |
--- SNIP ---

   Test Setup Configurations:

   There are five routers in the Test setup.  SHPE3, RR/P, MHPE1 and
   MHPE2 emulating a service provider network.  CE is a customer device
   connected to MHPE1 and MHPE2, it is configured with bridge domains in
   multiple vlans.  The router tester is connected to CE and SHPE3.The
   MHPE1 acts as DUT.The RT will be used as sender and receiver of
   traffic.The measurement will be taken in DUT.


WK: What exactly is a "Router Tester"? This term doesn't seem to be defined
anywhere -- is it just a traffic generator?
Please also fix the formatting above (missing spaces after periods) -
this is throughout the document.

   All routers except CE is configured with OSPF/IS-IS,LDP,MPLS,BGP with
   EVPN address family.

WK: All routers except CE are configured with. "Configured with"
doesn't really provide
  sufficient detail to allow repeatable testing.


   All routers except CE must have IBGP configured with RR acting as
   route reflector.

WK: iBGP. Also, expand acronyms.

   MHPE1,MHPE2,SHPE3 must be configured with "N" EVPN/PBB-EVPN instances
   depends up on the cases.

WK: s/MHPE1,MHPE2,SHPE3/MHPE1, MHPE2, SHPE3/ (and elsewhere)


   MHPE1 and MHEPE2 must be configured with ESI per vlan or ESI on IFD.

WK: What is an IFD?


   MHPE1 and MHEPE2 are running Single Active mode of EVPN.

   CE is acting as bridge configured with vlans that is configured on
   MHPE1,MHPE2,SHPE3.

   Depends up on the test traffic will be flowing uni directional or bi
   directional depends on the test performed.

WK: This is unparsable (and even if I could parse it I'm not sure what
it is related to)


   The above configuration will be serving as the base configuration for
   all test cases.

3.  Test Cases for EVPN Benchmarking

3.1.  Local MAC Learning

   Objective:

   To Record the time taken to learn the MAC address locally in DUT.

   Topology : Topology 1

   Procedure:

   The data plane MAC learning can be measured using the parameters
   defined in RFC 2889 section 5.8.  Send "X" unicast frames from CE to
   MHPE1(DUT) working in SA mode with "X" different source and
   destination address from RT.  The DUT must learn these "X" macs in
   data plane.

WK: MACs (and elsewhere)


   Measurement :

   Measure the time taken to learn "X" MACs in DUT evpn mac table.  The
   data plane measurement is taken by considering DUT as black box the
   range of X MAC is known from RT and the same must be learned in DUT,
   the time taken to learn "X" macs is measured.

WK: This is also largely unparsable.



   Repeat these test and plot the data.  The test is repeated for "N"
   times and the values are collected.  The mac learning time is
   calculated by averaging the values obtained from "N" samples.

   Mac learning in sec = (T1+T2+..Tn/N)

WK: Presumably (T1+T2+..Tn)/N ?


3.2.  Remote MAC Learning

   Objective:

   To Record the time taken to learn the remote macs.

   Topology : Topology 1

   Procedure:


   Send X frames with X different SA and DA to SHPE3 from RT.

WK: Please define / expand SA / DA. Also, SA conflicts with "SA EVPN
Terminologies SA Single-Active."

   SHPE3
   will advertise these locally learned macs to MHPE1 and MHPE2 via
   control plane.Measure the time taken to learn these X MACs from
   remote peer in DUT EVPN MAC address table.The DUT and MHPE2 are
   running SA mode.

   Measurement :

   Measure the time taken by the DUT to learn the "X" MACs in the data
   plane.Repeat these test and plot the data.The test is repeated for

WK: Repeat these tests...

   "N" times and the values are collected.The mac learning time is
   calculated by averaging the values obtained from "N" samples.

   Mac learning in sec = (T1+T2+..Tn/N)

WK: As above.  Also, what *exactly* do I report? This isn't (AFAICT)
"Mac learning in sec", it is (presumably) MAC Learning Rate, and I'd
*assume* that I report a line graph showing MACs learnt vs time? This
(and the other tests) need to be *much* more explicit about what
exactly is being reported.

--- SNIP ----

3.9.  ARP/ND Scale

   These tests are conducted to Record the scaling parameter of ARP/ND
   of the DUT.

   Objective:

   To Record the ARP/ND scale of the DUT.

   Topology : Topology 1

   Procedure:

   Send X arp/icmpv6 request from RT to DUT with different sender ip/
   ipv6 address to the same target gateway ip address.  Measure whether
   X MAC+IPv4 address/MAC+IPv6 address of the hosts are learned in DUT.

WK: I don't think I can run this test -- how fast do I send these? What
  *exactly* is the packet I'm sending (icmpv6 is not a packet, it's a protocol)
  Where do I send it? (to DUT? to the same target gateway IP? What
does that even mean?)
  The X MAC+IPv4 bit doesn't make sense.


   Measurement :

   The DUT must learn X MAC+IPV4/MAC+IPv6 and it must advertise the X
   MAC+IPV4/MAC+IPV6 to the remote router.

3.10.  Scaling of Services

   Objective:

   To measure the scale limit of DUT for EVPN.This is to measure the
   performance of DUT in scaling to "X" EVPN instances.

   Topology : Topology 1

   Procedure:
   The DUT,MHPE2 and SHPE3 are scaled to "N" EVI.Clear BGP neighbors of
   the DUT.  Once adjacency is established in the DUT.

WK: Fragment?

    Measure the
   routes received from MHPE2 and SHPE3 for "N" EVI in the DUT.

   Measurement :

   There should not be any loss of route types 1,2,3 and 4 in DUT.  DUT
   must relearn all type 1,2,3 and 4 from remote routers.  The DUT must
   be subjected to various values of N to find the optimal scale limit

WK: Do the first 2 sentences mean the same thing? Presumable I *do*
lose all the routes, but then I should recover them? Why is this being
compared to what I had before clearing, and not just the maximum /
number sent from the tester? What ratio should there be for types 1,
2, 3, 4? Is it important?


3.11.  Scale Convergence

   Objective:

   To measure the convergence time of DUT when the DUT is scaled with
   EVPN instance along with traffic.

   Topology : Topology 1

   Procedure:


   Scale N EVIs in DUT,SHPE3 and MHPE2.Send F frames to DUT from CE
   using traffic generator with X different SA and DA for N EVI's.  Send
   F frames from traffic generator to SHPE3 with X different SA and DA.
   There will be 2X number of MAC address will be learned in DUT EVPN
   MAC table.

WK: Doesn't parse.

   There is a bi directional traffic flow with F pps in each
   direction.  Then clear the BGP neighbors in the DUT.  Once the
   adjacency is restored in DUT.  Measure the time taken to learn 2X MAC
   address in DUT MAC table.

WK: Also doesn't parse.


   Measurement :

   The DUT must learn 2X MAC address.  Measure the time taken to learn
   2X MAC in DUT.  Repeat these test and plot the data.The test is
   repeated for "N" times and the values are collected.The convergence
   time is calculated by averaging the values obtained by "N" samples.

   Convergence time in sec = (T1+T2+..Tn/N)

3.12.  SOAK Test.

   Objective:

   This test is carried out to measure the stability of the DUT in a
   scaled environment with traffic over a period of time "T'".  In each
   interval "t1" the DUT CPU usage, memory usage are measured.  The DUT
   is checked for any crashes during this time period.

   Topology : Topology 1

   Procedure:


   Scale N EVI's in DUT,SHPE3 and MHPE2.Send F frames to DUT from CE
   using traffic generator with different X SA and DA for N EVI's.  Send
   F frames from traffic generator to SHPE3 with X different SA and DA.
   There will be 2X number of MAC address will be learned in DUT EVPN
   MAC table.  There is a bi directional traffic flow with F pps in each
   direction.  The DUT must run with traffic for 24 hours, every hour
   check for memory leak, CPU usage and crash.

   Measurement :

   Take the hourly reading of CPU, process memory.  There should not be
   any leak, crashes, CPU spikes.

WK: What all am I supposed to be reporting for this test? Can N be 1
and F be 1 and X be 1? I'm happy to do that and report no issues for
24 hours....


4.  Test Cases for PBB-EVPN Benchmarking

WK: ----- Are these the same as for "Test Cases for EVPN
Benchmarking"? If so, why are they repeated?

4.8.  High Availability

   Objective:

   To record traffic loss during routing engine failover.

   Topology : Topology 1

   Procedure:


   Send X frames to DUT with X different SA and DA from CE using the
   traffic generator.  Send X frames from traffic generator to SHPE3
   with X different SA and DA so that 2X MAC address will be Learned in
   DUT.  There is a bi directional traffic flow with X pps in each
   direction.  Then do a routing engine fail-over.

   Measurement :

   There should be 0 traffic loss which is the ideal case, No change in
   the DF role.  DUT should not withdraw any routes.Repeat the test "N"
   times and plot the data.The packet loss is calculated by averaging
   the values obtained from "N" samples.

   Packet loss in sec = (T1+T2+..Tn/N)

WK: What exactly am I reporting? "There should be 0 traffic loss which
is the ideal case" - am I reporting
the number of seconds with dropped packets? What if it *does* withdraw
routes? What do I report if I don't have multiple REs? Presumably if X
== 2 I can do this easily, but if X == 1e8 it's harder - but all I'm
reporting is "Packet loss".



--- SNIP ---

4.11.  Soak Test

   Objective:

   To measure the stability of the DUT in a scaled environment with
   traffic.

   Topology : Topology 1

   Procedure:

   Scale N PBB-EVI's in DUT,SHPE3 and MHPE2.Send F frames to DUT from CE
   using traffic generator with X different SA and DA for N EVI's.  Send
   F frames from traffic generator to SHPE3 with X different SA and DA.
   There will be 2X number of MAC address will be learned in DUT PBB-
   EVPN MAC table.  There is a bi directional traffic flow with F pps in
   Each direction.  The DUT must run with traffic for 24 hours, every
   hour check the memory leak, crashes.

   Measurement :

   Take the hourly reading of CPU process, memory usages.  There should
   not be any memory leak, crashes,CPU spikes.

WK: "any memory leak, crashes,CPU spikes" is not at all clear - e.g:
How are you defining a 'spike'? If the CPU jumps from 2% to 10% for 3
seconds, is that a spike? Many routers run a BGP reaper every N
seconds / minutes, which looks like a spike - is this an issue?

7.  Security Considerations

   There is no additional consideration from RFC 6192.

WK: I thought that all BMWG documents included standard boilerplate
for Security Considerations?

----------------

-- 
I don't think the execution is relevant when it was obviously a bad
idea in the first place.
This is like putting rabid weasels in your pants, and later expressing
regret at having chosen those particular rabid weasels and that pair
of pants.
   ---maf