high-fidelity switch models for sdn emulation

23
High-Fidelity Switch Models for SDN Emulation Danny Y. Huang Kenneth Yocum Alex C. Snoeren University of California, San Diego

Upload: lindsay

Post on 23-Feb-2016

47 views

Category:

Documents


0 download

DESCRIPTION

High-Fidelity Switch Models for SDN Emulation . Danny Y. Huang Kenneth Yocum Alex C. Snoeren University of California, San Diego. Buying OpenFlow Switches. Clients. Which one to buy?. HP Procurve Fulcrum Monaco Quanta LB4G. Buying OpenFlow Switches. 30% queries timed out. Clients. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: High-Fidelity Switch Models for SDN Emulation

High-Fidelity Switch Models for SDN Emulation

Danny Y. HuangKenneth YocumAlex C. Snoeren

University of California, San Diego

Page 2: High-Fidelity Switch Models for SDN Emulation

2

Which one to buy?

Buying OpenFlow SwitchesClients

HP Procurve

Fulcrum Monaco

Quanta LB4G

Page 3: High-Fidelity Switch Models for SDN Emulation

3

Buying OpenFlow SwitchesClients

30% queries timed out

Page 4: High-Fidelity Switch Models for SDN Emulation

4

Buying OpenFlow SwitchesClients

Page 5: High-Fidelity Switch Models for SDN Emulation

5

Buying OpenFlow SwitchesClients

Same topology. Same workload.

Different performance!

Page 6: High-Fidelity Switch Models for SDN Emulation

6

Buying OpenFlow SwitchesPerformance Variations

x 1,000 =

Clients

How to predict the performance?

Which one to buy?

Page 7: High-Fidelity Switch Models for SDN Emulation

7

Emulating OpenFlow Networks

Controller

OpenFlow

Data Traffic Data Plane

Simulators / Emulators

Control PlaneOpen vSwitch

Mininet

OVS itself would also introduce performance variations.

Page 8: High-Fidelity Switch Models for SDN Emulation

8

Problem

Hard to predict OpenFlow network’s performance:

• OpenFlow Switches are different.

• Existing emulation framework is not good enough.

Goal

To predict performance with realism:

• To design an emulator that captures vendor variations.

• To measure these variations in the control plane.

Page 9: High-Fidelity Switch Models for SDN Emulation

9

Variations across Vendors

Controller

OpenFlowProtocol

Data Traffic

Differences in Control Plane

Flow table size

Flow management policies

Switch CPU

etc

Focus on CPU’s effect on control-path delays.

Page 10: High-Fidelity Switch Models for SDN Emulation

10

Control-Path Delays

Hardware OpenFlow

Switch

Controller (POX)

Application Workload

Ingress

Packet-inEvents

Flow-modEvents

EgressData plane traffic

Ingress Delay

Egress Delay

TCAM

CPU

Disproportionately affects short flows.

Page 11: High-Fidelity Switch Models for SDN Emulation

11

Measure Control-Path Delays

Server

Eth 0

Eth 1

Eth 2

OF Switch

Data Plane

Control PlaneController

Clients

HP Procurve

Fulcrum Monaco

Quanta LB4G

Open vSwitch (OVS)

Installs rules for every new flow

Queries for a 64-byte value every 50 ms. We measure the query time, ingress & egress delays.

Page 12: High-Fidelity Switch Models for SDN Emulation

12

Measure Redis Query Times

Query completion times for Redis clients (ms)

Page 13: High-Fidelity Switch Models for SDN Emulation

13

Measure Ingress DelayOVS faster than

the others.

Page 14: High-Fidelity Switch Models for SDN Emulation

14

Measure Egress Delay

Slow down the ingress and egress delays on OVS to emulate the hardware

OVS: almost no delays!

Page 15: High-Fidelity Switch Models for SDN Emulation

15

Implementing the Emulator

Open vSwitch (OVS)

Application Workload

Ingress

Packet-in Events

Data plane traffic

Controller (POX)

Flow-mod Events

Egress

To slow down control traffic

Page 16: High-Fidelity Switch Models for SDN Emulation

16

Emulator Proxy

Open vSwitch (OVS)

Emulator Proxy

Application Workload

Ingress

Packet-in Events

Data plane traffic

Controller (POX)Packet-in Events

(Delayed)

Flow-mod Events(Delayed)

Egress

Flow-mod Events Physical OF Switch≈

Page 17: High-Fidelity Switch Models for SDN Emulation

17

Evaluation

Query completion time for Redis clients (ms)

CDF

HP

OVS

Hardware

Emulated

Page 18: High-Fidelity Switch Models for SDN Emulation

18

Query completion time for Redis clients (ms)

Evaluation

CDF

Query completion time for Redis clients (ms)

HP

Monaco

OVS

Hardware

Emulated

• We emulate the ingress and egress delays only.

• Reasonble approximation.

Query completion time for Redis clients (ms)

Quanta

Page 19: High-Fidelity Switch Models for SDN Emulation

19

Summary

• Hard to predict performance due to vendor variations.

• We designed an emulator for control-path delays.

• Simple, but achieves reasonable approximation.

Future Work

• Increase realism. Capture more artifacts.

• Expand workload coverage. Automate switch measurements.

• Capture interactions among multiple switches.

Acknowledgements Marco Canini and Dan Levin (TU Berlin)George Porter (UC San Diego)

Page 20: High-Fidelity Switch Models for SDN Emulation

20

Thank you!

Page 21: High-Fidelity Switch Models for SDN Emulation

21

Inverse Transform Sampling

• Goal: Emulate switch X, which introduces ingress delay t with probability p.

• Algorithm:1. Measure the delay distributions of OVS and X. Make

them into tables.2. Measure how much OVS has delayed. Call this tOVS.3. Look up tOVS from the OVS table. This returns probability

p.4. Look up p from X’s table. This returns delay tX.5. Introduce delay (tX - tOVS).

Page 22: High-Fidelity Switch Models for SDN Emulation

22

Evaluation (QQ Plots)

HP Monaco Quanta

Tim

e on

Em

ulat

or (m

s)

Query completion time on hardware switches (ms)

Page 23: High-Fidelity Switch Models for SDN Emulation

23

Time Dilation