overviejmanbal/paper/acacia-poster.pdf · 2018. 3. 29. · lte connection to internet ① open a ci...
TRANSCRIPT
ACACIA – Context-aware Edge Computing for Continuous Interactive Applications over Mobile Networks
ACACIA Design
Overview NSF award: #1305384 #1343713
Motivation • Continuous Interactive (CI) mobile apps
emerging, e.g., AR, VR, and face recognition
• Many use cases – Retail, stadiums, tourism, museums etc
• Characteristics of CI apps
– Highly responsive – Compute intensive
Junguk Cho, Karthikeyan Sundaresan, Rajesh Mahindra, Sampath Rangarajan and Jacobus Van der Merwe
Open Question: How should the three entities (user, network, and application) be jointly orchestrated and combined in a service abstraction to deliver low latencies and enable CI application in mobile network?
ACACIA Demo Scenario LTE Connection to Internet ① Open a CI app (User context) ② Bind the CI app to ACACIA Device Manager ③ Send user context to MRS from ACACIA Device Manager ④ Create policy rules for the user’s CI app ⑤ Setup a dedicated bearer in network for the user’s CI App through
GW-Cs, MME, eNodeb and LTE modem in mobile device ⑥ Setup flow rules in local S&PGW-U for the dedicated bearer ⑦ Running the CI app with a CI server
MEC RegistrationServer (MRS)
CI Server PCEF
SGW&PGW-C
LTE Modem
Dedicated Bearer (MEC)
Default Bearer (Internet)
MEC
ACACIA Device
Manager
EPC
CI App
UE
Open vSwitch(Local S&PGW-U)
UL TF
Tfo
r MEC
UL TF
T for
In
terne
t
MME PCRF
Open vSwitch(S&PGW-U)
eNodeBs
Ryu SDN ControllerOpenFlow
3GPP Control Plane
3
4
5
6
71
2
0
0.5
1
1.5
2
2.5
0 10 20 30 40 50 60 70 80 90 100
Netw
ork
late
ncy
(se
c)
Background traffic (Mpbs)
Conventional EPC
EPC with MEC
ACACIA
MEC Isolation of CI Traffic Control Overhead RTT Reduction
Always-on for an extra bearer LTE Radio Promotion Event (Release and re-establish) - One event requires 15 control msgs (SCTP (7), GTPv2 (4), OF(4))
Per day Average Worst # of times 929 7200 Data (MB) 2.58 20
0
0.2
0.4
0.6
0.8
1
13 14 15 16 17 18
CD
F
RTT (ms)
QCI 5QCI 6QCI 7QCI 8QCI 9
App Optimization Matching time
0 0.5
1 1.5
2 2.5
1 2 4 8
Matc
hing t
ime (
sec)
The number of clients
ACACIArxPowerNaive
DB size :105 objects Image resolution : 920x720
Current Solutions • Computation offloading to cloud servers
– No consideration for mobile network complexities
• Mobile Edge Computing (MEC) over mobile network
– Need to modify current mobile network infrastructure and establish new specifications
• Context discovery & user’s localization – Significantly help optimize CI application
ACACIA - Service abstraction framework - Adapts a holistic end-to-end approach to
enabling low latency CI services over existing mobile networks
- Leverage client context information through LTE-direct to optimize both network and application processing
ACACIA Design (i) User context discovery
- LTE-direct : Pub-sub mechanism - Publish “service” availability - Subscribe to “interests”
(ii) Context-aware traffic redirection - Request a connectivity between a CI app and closest CI server (on demand)
(iii) ACACIA Mobile Edge Network - Use LTE/EPC QoS bearer framework with
SDN and NFV GW-Us (standards-compliant MEC) (iv) Context-aware application
optimization - ACACIA device manager provides
location information to the CI server
ACACIA Use Case
LTE RTT (~70 ms)
Retail Shop
AR Server
Computer)Sec,on)
Food)Sec,on)
Salesman
Salesman
“Interest” in laptop
Retail)App)(AR))
“Service” Match
“Service” Non-Match
1"
1"
Customer
2"3"4"
Send: User context info & screen view Recv: Info relevant to the scene ( e.g., price, reviews)
4"
ACACIA Architecture
eNodeB
ACACIAdevice
manager
ACACIAdevice
manager
EPCfunctions
SDN/NFV EPCfunctions
Sub/CIApp
CI Server
Mobile device
Mobile device
Mobile Edge Clouds
Core Mobile Network
User context discovery
Context-aware traffic redirection andNFV-enabled offloading
Context-aware application optimization
LTE-direct communicationNetwork Control
Network Data
Location information
i
ii
iii
iv
Radio Access Network
Pub App
Solutions proposed and studied in standalone manner