Leveraging OpenFlow for Resource Placement of Virtual DesktopsProject Team: Prasad Calyam, Ph.D. pcalyam@osc.edu,

Sudharsan Rajagopalan, Arun Selvadhurai, Alex Berryman, Saravanan Mohan, Prof. Rajiv Ramnath

GEC15 Plenary Session TalkOctober 2012

Research Sponsors: NSF (CNS-1050225, CNS-1205658), VMware

Topics of Discussion

• VDCloud-GENI Experiment Context

• ‘VDC-Sim’: Virtual Desktop Cloud Simulator– Research use cases– Education use cases

• VDCloud Experiment Demonstration– GENI Slice setup– OpenFlow integration– VDC-Sim results ‘with’ and ‘without’ load balancing

2

Virtual Desktop Clouds (DaaS)

“Brain of the Cloud”

3

Roller Coaster Track Design• Understand energy transfer (potential to kinetic)

– Build cool coasters, study tsunamis – similar science• Optimal design: hills, bigger loops, more cars, safe stop

4Credit: National Geographic, The Jason Project

Roller Coaster Test

5

Roller Coaster Performance

6

User Experience

Resource Allocation Net-utility

VMLab-GENI Experiment Context“VDC-Sim” Provisioning and Placement

GENI Slice Testbed 7

• VDC-Sim → GENI • Design & Development →

Validation and design tuning• Large-scale simulations →

Cloud deployment experiments

VDC Research “Big Picture”

1. VDBench: Thin-client Performance Benchmarking

2. Deriving Application Behavior Profiles

3. Utility-directed Resource Allocation Model (U-RAM) for VD Provisioning

4. Local/Global Distributed Optimization for VD Placement

8

VDC-Sim Features

‘Run Simulation’(Offline)

‘Run Experiment’(In GENI)

Net-utility per experiment run

Resource allocation ofthin-clients to data centers

9

VDC-Sim Demo-1

Net-utility = 21.64

Resource allocation ofthin-clients to data centers

10

VD Requests Load = 100; No Fault Occurrences

VDC-Sim Demo-2

Net-utility = 20.44

Resource allocation ofthin-clients to data centers

11

VD Requests Load = 100; Fault Occurrence Levels = 50

Use Cases

• Research– Plug-in new provisioning and placement schemes– Study cloud dynamics to see how they affect net-utility

• Education– Explore server-side adaptation

• E.g., write a macro script to reduce user interaction round-trips for control actions during network health bottlenecks

– Explore client-side adaptation • E.g., select thin-client encodings that delivers best QoE for

different user groups – knowledge worker vs. designer/artist

12

Marker Packet Header Format

13

OpenFlow Switch

OpenFlow Controller

SmartThin-client

Virtual Desktop

Join OpenFlow network

Install flow rules for marker packets

Send marker packet to request virtual desktop

Recognize and punt the marker packet

Parse marker packet and install client/server flows

Access virtual desktop applications

Flow Setup Sequence Diagram

1

2

3

4

5

6

14

VDCloud Experiment w/o Load-Balancing

15

VDCloud Experiment w/ Load-Balancing

16

OpenFlow Switch

Client In Por

t

Out Port

SUNNW PG48 50 51

SUNNW PG49 50 51

ATLANTA PG46 52 52

ATLANTA PG47 52 52

ATLANTA PG46 20 52

ATLANTA PG47 20 52

Application Cross-Traffic0

5

10

15

20

0.21

15.36

Demonstration

17

Route setupStep-1 Cross-traffic ImpactStep-2 Load-balancing

ImprovementStep-3

OpenFlowSwitch

Client In Por

t

Out Por

tATLA PG46 20 52

ATLA PG47 20 52

OpenFlow Switch

Client In Por

t

Out Por

tATLANTA PG46 20 52

ATLANTA PG47 20 52

SUNNW PG48 50 52

SUNNW PG49 50 52

Video runs smooth, GUI applications are responsive

Video freezes, disconnects, GUI applications are not responsive

Video runs smooth, GUI applications are responsive

Bandwidth Consumed (Mbytes/s)

Application Cross-Traffic0

5

10

15

20

4.45

14.8

Application Cross-Traffic0

5

10

15

20

4.6

0

Thank you for your attention!

18