mobiscud:afastmovingpersonal …jmanbal/paper/mobiscud_slide.pdfmobiscud:afastmovingpersonal...

MobiScud: A Fast Moving Personal Cloud in the Mobile Network

Kaiqiang Wang , Minwei Shen Junguk Cho , Jacobus Van der Merwe

Arijit Banerjee, Kirk Webb

2/6/16 1

Various Wearable Devices

2/6/16 2

With Wearable Device

•  The Google glass keeps doing facial recogniMon

or analysis.

•  The smart wristband detects and records my heartbeats and blood pressure, giving me advices.

•  More on the way… 2/6/16 3

ApplicaMon Requirements •  CharacterisMcs of these applicaMons

–  Highly responsive : should get results in real-‐Mme (AR, face recogniMon) –  Intensive computaMon : hard to run only on mobile –  Need storage to keep and analyze user-‐related informaMon

Need computa3on offloading to cloud server everywhere and any3me

However, the first priority of wearable devices is not computaMon and storage, but be[er portability

2/6/16 4

With Cloud & Mobile Network

•  Cloud server –  Provide compuMng resource & storage for wearable & mobile devices

•  Stable connecMvity –  LTE is everywhere and always-‐on (Macro, small cells) –  LTE is more stable than WiFi since it runs on license spectrum

However, cloud server far from users and centralized GWs (hierarchical rou3ng) in Core network result in unpredictable and long delay

Offset benefit from computa3on offloading due to network delay

Cloudlet* approach on mobile network Cloudlet* -‐ The case for vm-‐based cloudlets in mobile compu8ng, Satyanarayanan et al. 2/6/16 5

Cloudlet

Source: The case for vm-‐based cloudlets in mobile compu8ng , Satyanarayanan, et al.

2/6/16 6

In Cloudlet, they designed Cloudlet based on WiFi network

WiFi AP

We tried to use this Cloudlet approach in mobile network

•  Deployment –  Put cloudlets with or close to eNodeB –  ModificaMon of exisMng mobile network is difficult

•  Rigid architecture and complex protocol in mobile network

•  Mobility –  As an user moves, iniMal cloudlet for the user can be gradually far from the user ( e.g., Handover )

–  Considering target applicaMons are highly latency-‐sensiMve, increasing small latency drops user’s QoS a lot

Challenges of Cloudlet on Mobile Network for Low Latency

2/6/16 7

Design Requirements •  Highly distributed cloudlet plaform

•  Deploy cloudlet approach without modificaMon of current protocol or mobile network

•  Guarantee constantly short latency regardless of mobility

2/6/16 8

Proposed Architecture : MobiScud •  Highly distributed cloudlet plaform –  Cloudlet with or close to Radio Access Network (RAN)

•  Apply cloudlet approach without modificaMon of current protocol or mobile network –  SoMware-‐Define Networking (SDN)

•  Guarantee constantly short latency regardless of mobility –  Personal cloudlet keeps following the owner –  Combine handover procedure & cloudlet migra3ons

2/6/16 9

MobiScud Architecture

eNB

Cloud

SGW PGW

MME

Internet

eNB

Cloud

SDN

SDN

MobiScud

MobiScud

2/6/16 10

LTE/EPC Architecture


eNB

Cloud

SGW PGW

MME

Internet

eNB

Cloud

SDN

SDN

MobiScud

MobiScud

2/6/16 11


MobiScud components


eNB

Cloud

SGW PGW

MME

Internet

eNB

Cloud

SDN

SDN

MobiScud

MobiScud

2/6/16 12


MobiScud components

•  MobiScud Monitor –  Monitor the control plane messages ( e.g., a[achment and handover event ) –  Extract necessary informaMon from them for future flow rules


eNB

Cloud

SGW PGW

MME

Internet

eNB

Cloud

SDN

SDN

MobiScud

MobiScud

2/6/16 13


MobiScud components

•  MobiScud Controller -‐ Orchestrate SDN substrate and Cloud plaWorm –  Construct flow rules with extracted informaMon from Monitor –  Manage cloud plaform to create and migrate the personal VMs –  Exchange control message between MobiScud controllers to support handover


eNB

Cloud

SGW PGW

MME

Internet

eNB

Cloud

SDN

SDN

MobiScud

MobiScud

2/6/16 14


MobiScud components

•  SMORE* SDN –  SelecMve traffic offloading evaluaMon based on flow-‐entries : Cloud or Internet –  GTP Tunnel en/decapsulaMon

SMORE*: soCware-‐defined networking mobile offloading architecture, J Cho et al.

MobiScud Workflow for Handover

eNB1

SGW PGW

MME

Internet

MobiScud1

1

Cloud1

SDN1

2/6/16 15

AMer user aYachment Traffic for Cloud Traffic for Internet


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2 Cloud2

Cloud1

SDN2

User start moving, Handover procedure starts

2/6/16 16

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

2

SDN2

2/6/16 17

Cloud2

Cloud1

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

3

Cloud1

Cloud2

SDN2

2/6/16 18

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

4

Cloud1

Cloud2

SDN2

2/6/16 19

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

5 Cloud1

Cloud2 Request VM migra3on

2/6/16 20

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

6

Start VM migra3on

2/6/16 21

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

6

Handover completes

2/6/16 22

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

6

S3ll use VM in cloud 1

2/6/16 23

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

6

S3ll use VM in cloud 1

2/6/16 24

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

MobiScud1

MobiScud2

Cloud1

Cloud2

7

AMer finishing VM migra3on Modify flow rules

2/6/16 25

SDN2

SDN1

eNB2


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

8

Use VM in Cloud2

2/6/16 26

SDN2

SDN1


eNB1

SGW PGW

MME

Internet

eNB2

MobiScud1

MobiScud2

Cloud1

Cloud2

9

2/6/16 27

SDN2

SDN1

ImplementaMon

•  PhantomNet testbed –  hYp://phantomnet.org/ – OpenEPC LTE/EPC somware

•  MobiScud SDN & Controller –  SMORE SDN & Ryu controller

•  Personal cloudlet –  Xen

2/6/16 28

EvaluaMon

•  End-‐to-‐End RTT improvement according to the level of cloud distribu3on –  Cloud in Internet –  Cloud in intermediate locaMon between eNB and Core network (MTSO) –  MobiScud without VM MigraMon –  MobiScud with VM MigraMon

•  We use ping for evaluaMon 2/6/16 29

EvaluaMon

•  End-‐to-‐End RTT improvement according to the level of cloud distribu3on –  Cloud in Internet –  Cloud in intermediate locaMon between eNB and Core network (MTSO) –  MobiScud without VM MigraMon –  MobiScud with VM MigraMon


EvaluaMon

•  End-‐to-‐End RTT improvement according to the level of cloud distribu3on –  Cloud in Internet –  Cloud in intermediate loca3on between eNB and Core network (MTSO) –  MobiScud without VM MigraMon –  MobiScud with VM MigraMon


EvaluaMon

•  End-‐to-‐End RTT improvement according to the level of cloud distribu3on –  Cloud in Internet –  Cloud in intermediate locaMon between eNB and Core network (MTSO) –  MobiScud without VM Migra3on –  MobiScud with VM MigraMon


EvaluaMon

•  End-‐to-‐End RTT improvement according to the level of cloud distribu3on –  Cloud in Internet –  Cloud in intermediate locaMon between eNB and Core network (MTSO) –  MobiScud without VM MigraMon –  MobiScud with VM Migra3on


RTT of Different Cases

2/6/16 34

Constant ~80 ms latency

Constant ~20 ms latency


First Handover Second Handover

2/6/16 35


First Handover Second Handover

2/6/16 36

Conclusion •  We presented MobiScud architectue to realize cloudlet in mobile network for low latency applicaMons –  No modificaMon of exisMng LTE/EPC mobile network. –  Keep following personal cloudlet even handover

•  Prototype realizaMon of MobiScud architecture in PhantomNet LTE/EPC testbed.

2/6/16 37

Conclusion •  We presented MobiScud architecture to realize cloudlet in mobile network for low latency applicaMons –  No modificaMon of exisMng LTE/EPC mobile network. –  Keep following personal cloudlet even handover

•  Prototype realizaMon of MobiScud architecture in PhantomNet LTE/EPC testbed.

Visit phantomnet.org 2/6/16 38

mobiscud:*a*fastmoving*personal* …jmanbal/paper/mobiscud_slide.pdfmobiscud:*a*fastmoving*personal*...

Documents

mobiscud:afastmovingpersonal …jmanbal/paper/mobiscud_slide.pdfmobiscud:afastmovingpersonal...