1. Rethinking LTE Network Function Virtualization
Muhammad Taqi Raza*¶, Dongho Kim§, Kyu-Han Kim§, Songwu Lu* and Mario Gerla*
* Computer Science Department, UCLA
§ Hewlett Packard Labs
¶ Student and corresponding author.

2. Network Function Virtualization (NFV)
NFV replaces dedicated network functions (NFs) with software running on commercial commodity servers.
Dedicated
Intrusion
Detection
Dedicated
Firewall
Dedicated
Load
Balancer
Commodity
Intrusion
Detection
Commodity
Firewall
Commodity
Load
Balancer
Advantages
Low Cost
Reduces capital and operational expenditures
Flexibility
Network functions can be chained dynamically
Scalability
Network functions are quickly scaled up and down

3. LTE – NFV: A Leading Use Case of NFV
A large variety of proprietary LTE NFs negatively impact efficiency
 NFV moves away from propriety boxes and improves efficiency
Launching services requires another variety of box to be integrated
 Launching service in NFV is as easy as upgrading a software
Operations are slow and expensive
 NFV provides automated and agile solutions to scale network services

4. Virtualization Platform
Traditional Way of NFV
Multiple NF instances are created to meet greater subscribers demands
NFV dynamically selects NF for packet processing:
Dynamically selects NF instance, and
Dynamically routes the network packets
Virtualization Platform

5. Problem: Traditional Way of NFV in LTE
LTE NFs are not Internet Middleboxes
Subscriber Runtime NF Delay Reroute
Affinity Selection Sensitive Flows
Internet Middleboxes
LTE NFs

6. Achieving: LTE Way of NFV
Different LTE events require different treatment
 Treat LTE events on their merit
LTE signaling packets drive performance
 Prioritize delay sensitive signaling messages
 Parallelize LTE signaling messages
Must meet standardized conformance and interoperability requirements
 Ensure in-order execution of signaling messages
LTE NFs are logically separated
 Combine logic of alike NFs

7. LTE Way of NFV: How We Do ? +
Logic based NF decomposition instead of instance based NF decomposition
Instance based NFV
(traditional way)
LTE
Serving
Gateway
Packet
Dedicated Commodity
boxes boxes
Logic based NFV (LTE way)
Dedicated Commodity
boxes boxes
Event 1
Logic
LTE
Serving
Gateway
Packet +
Event 2
Logic

8. LTE Way of NFV: How We Do ?
A number of signaling messages exchange between distributed LTE NFs
Logic based NFV (LTE way)
MME
VNF
SGW
VNF
PGW
VNF
Event
Input
LTE Core Network
Chaining for normal event
MME: Mobility Management Entity
VNF: Virtualized NF
SGW: Serving Gateway
PGW: Packet Gateway

9. LTE Way of NFV: How We Do ?
Delegating LTE event execution to Fat-Proxy
Mission critical events are delegated to Fat-Proxy
Logic based NFV (LTE way)
MME
VNF
SGW
VNF
PGW
VNF
HoM VNF
SP VNF
Check
Event
Input
LTE Core Network
Fat-Proxy Tier
Chaining for normal event
MME: Mobility Management Entity
VNF: Virtualized NF
SGW: Serving Gateway
PGW: Packet Gateway
HoM: Handover Management event
SP: Service provisioning event
Chaining for mission-critical event

10. LTE Way of NFV: How We Do ?
Parallelizing mutual exclusive signaling messages
MME (LTE NF) communication with base station and serving gateway is mutually exclusive
Logic based NFV (LTE way)
MME Core Function
GTP
S1-AP
SGW
LTE base station
Mobility Management
Entity (MME)
GTP: GPRS Tunneling protocol
S1-AP: S1 Application Protocol

11. Challenge 1: Functional Decomposition
Fat-Proxy implements event specific logic
The NF source code is shared among a number of LTE events.
Challenge 1-1: Determine what software functions are shared among multiple events ?
Challenge 1-2: Determine indirect dependencies

12. Solution C1-1: Functional Decomposition
Decomposing event specific functions from the source code
Generating function call graph to determine functional dependency
Different software functions chain differently based on the event logic:
Modify Bearer Req.
Create Session Req.
Location Update procedure
Modify Bearer Resp.
Create Session Resp.
Handover procedure

13. Solution C1-2: Functional Decomposition
Global variable usage as a reason for functional dependency
Example below:
Variables ‘bearer’ and ‘imsi’ values are modified by some other functions
PathSwitchRequest() is dependent on MapIMSI.find()
PathSwitchRequest() is dependent on GetEPSBearer()
function PathSwitchRequest (enbUeS1Id, mmeUeS1Id) {
// get IP address of UE by removing header …
// find corresponding UeInfo address
imsi = MapIMSI.find (ueAddr);
//get UE corresponding eps bearer
bearer = GetEPSBearer(); }

14. Challenge 2: Event Logic Extraction
Extract critical event execution logic from LTE core NFs
Combine extracted logic as that event’s Fat-Proxy
Challenge 2-1: Resolving logic and data dependencies
Challenge 2-2: Resolving event execution dependencies

15. Solution 2-1: Event Logic Extraction
Resolving event execution dependencies
Identifying logic dependencies through Common Subgraph Isomorphism
Offline process through improved back-tracking algorithm
Handover procedure
Location Update procedure
Handover Required
Authentication Req.
Create Session Req.
Modify Bearer Req./Resp
Authentication Resp.
Create Session Resp.
Context
Ack.
FWD Relocation
Create Session Req.
Create Session Resp.
Dependent
Modify Bearer Req.

16. Solution 2-2: Event Logic Extraction
Resolving event execution dependencies
Device Powers on
Location Req. initiated
Locat. Upd. accepted/ rejected
Locat. Upd. requested
Registered initiated
Registered
Attach accepted
Srvc Req. initiated
Service Req. accepted/failed
Service Req initiated

17. Challenge 3: Logic-based Partitioning
Speed-up event execution by executing some messages in parallel
Challenge 3-1: By design serial execution of signaling messages at LTE core
GTP: GPRS Tunneling protocol
S1-AP: S1 Application Protocol

18. Solution 3-1: Logic-based Partitioning
Partition the mutually exclusive logic of different protocols
Only execute those messages in parallel which:
Belong to two different protocols
Are mutual exclusive
Base station
Source-MME
Target-MME
Source-SGW
Target-SGW
Handover required
FWD Allocation
Handover Request
Create Session Req.
Handover ACK
Session Response
FWD Allocation Resp.
Handover Command
Create Indirect data FWD tunnel
eNodeB Status
Transfer
FWD tunnel resp.
GTP: GPRS Tunneling protocol
S1-AP: S1 Application Protocol
Messages executed in sequence
S1AP Protocol messages executed in parallel
GTP Protocol messages executed in parallel

19. Implementation and Evaluation
LTE base station: (nanoLTE Access Point)
LTE Core network: OpenEPC software platform
Device: Samsung S6 smartphones
Virtualization: VmWare’s vSpehere
Intel Xeon E v3 processors at 2.3Ghz

20. Event Execution Time
Event execution being local to Fat-Proxy instance speeds-up different events
Paging event: Most packets are exchanged between LTE core and LTE base station
Not much improvement in Paging event through Fat-Proxy

21. Event Execution Time Event execution is diverted to Fat-Proxy
Less number of packets exchange at LTE core
Service Request event: Bearer modifications at actual SGW and PGW which relatively increases LTE core signaling
Not much improvement in Paging event through Fat-Proxy

22. Conclusion
Made the first effort in providing logic based decomposition in Virtualized EPC
Design leverages LTE domain knowledge to extract the event logic
By using domain-specific knowledge in LTE, our design does not require any LTE standard violation
We seek for plug and play solution to work with any carrier network
In future work, we will focus on service availability (fault tolerance), LTE core platform (for centralized control) and security issues in LTE–NFV