MoM
- 1. Testbed
- 2. Discuss Results
  - 3. SW2SW discussion

MoM

ZD, MK, XTD, PE, Tobi D

Start 10:40: setup testbed

1. Testbed

Manzoor introduces testbed components
- ZD: documentation of the developed software (wifi monitoring tools)
- ZD: controller should really send flow-rules for dummy Packet-Ins to OVS. To be realistic.
- ZD: the wifi tools: server, client can be turned into contributions.
  - MK: the tools are useful also in connection with wireless port virtualization, so that the controller sees each single logical wifi port.
  - ZD: impressed.
  - MK: we can use real streaming for demo
  - ZD: we must not push if it's complicated. But do if it's easy.
  - ZD: suggests make further use of the tools, details how they works, generate report.. contribution.
  - MK: propagration delay
  - ZD: More OVS, devices to generate load. Generate load form other devices, see how the controller is affected.
  2. Discuss Results
TD shows experiment results.
- MK: for now we can not see impact of number of intermediary SWs on delay because only 1 sw is used now.
- ZD: Why physical?
- MK, PE: behavior of OVS is not clear.
- MK: mix phy and virt SWs (10- 20 pieces).
- ZD: one idea here. classifying user packages using some bits in the middle. Destination is rewritten by core network (…) to fast forward packets to dest and rewrite back by the access points. –
  
  > Delay is 2ms. The experiment is with mininet, 100 nodes, big server.
  - ZD: our scale is still 1 sec. We need to understand the different components. To explain the big jump in results.
  - Propagation delay:
  - PE: unknown b/w AP and Ctrl due to DAI network.
  - ZD: SW-Ctrl should go in-band for realistic setup. Mininet has direct conn with ctrl. Propagation part could be removed if it's not realistic?
  - MK: then we only have Ctrl load with simple testbed.
  - MK: we aim at a testbed with analytical function for propagation delay. This will be good contributions.
  - ZD: Understand all components. Maybe the case where ctrl acts ahead of propagated packets.
  - ZD: Relate contribution to mobile network architecture, thou we could not play around with real LTE, 5G testbed.
  =

3. SW2SW discussion

MK: we want to integrate SW2SW to the demonstrator. But not for now, only after the testbed get more complex. ZD mentioned are relation with IPR.

ZD: not really have this relation in mind, but for other part…

MK, ToD: show SW2SW slides of 07-04-2016

ZD: the gain?

MK: this reduces interactions with controller

ZD: we need to see what case it gains / or not.

ZD: how about using the approach aggregation approach above to reduce recofiguration of downstream SWs? CO rewrite MH packets and downstream SWs treat them as their current flows. (ZD share slides).

ZD: why TTL for trigger packets? It's Ok with testbed environments.

ZD shows slides of aggregation ideas. We may find the way to combine with DAI SW2SW idea.

Controller involves in the preconfiguration. As the current state.
ZD: SW2SW should be more intellig. The next upstream (inter.) sw accepts the trigger(?) as if it came from the controller.

ZD invited Khalil, he likes the demo.

Demo paper discussion:

ZD: demo paper. What impact the MM most? delay component?.
K: how to calculate/formulate delay?
MK: We observe from testbed and study the analytical function. We test with more phy SWs to see the impact.
ZD: Divide total delay to componenets is resonable. K agreed. We need to study the components.
MK: We formulate functions for the components based on study on testbed. See what component is important.
MK: It is important for us / Huawei to document the testbed.
ZD: put the wifi monitoring tools.
ZD: is OpneWRT / other solution for AP? It makes sense to use optimized hardware.
PE: not aware of the current support for OVS.
ZD: if no option is avail. there is room for contribution.
MK: Aetherflow is the closest idea but not implementation avail.
K: Virtualize APs on Laptops: separate AP and OVS functions.
ZD: Integrate OVS + OpenWRT is a nice contirbution for people to use.
ZD: Keep the demo paper as living document with contribution from both sides.

Discuss result with Khalil

K: Nope.

Table of Contents

MoM

1. Testbed

2. Discuss Results

3. SW2SW discussion