2. Telco Clouds: Modelling and Simulation
Jakub Krzywda1, William Tärneberg2,
P-O Östberg1, Maria Kihl2, Erik Elmroth1
1 – Umeå University
2 – Lund University
MY NAME IS
OTHER AUTHORS are…
During this presentation I would like to tell you about our APPROACH to MODELING and SIMULATION of telco clouds.

3. Multi-Model
IN THIS PAPER: We propose a MULTI model ->

4. … that combines: Telecommunication Cloud computing
that combines telecommunication and cloud computing infrastructures ->
Telecommunication
Cloud computing

5. … to simulate the influence of:
to simulate the influence of users mobility and infrastructure setup ->
Mobility
Infrastructure Setup

6. … on:
on the system performance and costs.
Performance
Costs

7. Computation Offloading
Mobile Devices
high requirements on Quality of Service
limited battery lifetime
less powerful resources
To give some MOTIVATIONS lets take a look on following example:
Mobile phones – NOWADAYS not only making CALLS but also for RUNNING APPLICATIONS
Applications – RICH USER INTERFACES and high DEMANDS on computational resources and QoS
However… limited battery, less powerful resources than in desktops
TO ENABLE running this kind of applications we can use COMPUTATION OFFLOADING

8. Computation Offloading
Computations are offloaded from # mobile phones to # data centers: DATA ARE FIRST SENT to # radio base stations and then via # Internet to data centers, processed there and results are sent back

9. Other motivations
But computation offloading is NOT THE ONLY motivation
SELF DRIVING CARS
INTERNET OF THINGS with MILLIONS of SENSORS sending data through internet

10. Current Situation Telecommunication Infrastructure
HOWEVER in the current situation # Telecommunication infrastructure – such as radio base station and # cloud computing infrastructure such as data centers are # SEPERATED and MANAGED INDEPENDENTLY
Cloud Computing Infrastructure

11. Issues Latency Hardware costs Network Congestion
That leads to several issues. For example,
# Latency due to distance
# with EXPLOSION of Internet of Things - millions of SENSORS sending data for processing/storing – # may cause network congestion
# todays telecommunication hardware is EXPENSIVE
Network Congestion

12. Introducing Telco Cloud
To solve this Telco Cloud concept WAS PROPOSED

13. Current Situation Merging Telecommunication and Cloud Infrastructures
CURRENTLY Telcommunication and cloud infrastructures are SEPARATED
Telco Cloud OPENS THAT BARRIER and MARGES both infrastructures

14. Telco Cloud Introducing Proximal Data Centers
by PLACING additional PROXIMAL Data Centers at the EDGE of MOBILE NETWORK (closer to users)

15. Benefits Reduce Latency Decrease costs Prevent Network Congestion
Merging Telco and Cloud infrastructures to
increase QoS (e.g. by reducing latency)
prevent network congestion (e.g. by FLITERING some data and PROCESSING them LOCALLY)
decrease costs of operation (e.g. by virtualizing some parts of radio base stations and moving them into the cloud)
Prevent Network Congestion

16. Telco Cloud Dynamics Applications Mobility Topology DC Capacity
Workload
Objectives
Setup
Applications
Mobility
Topology
DC Capacity
Service placement
Quality of Service
Costs
BEFORE MODELLING and specifying parameters IN THIS PAPER WE analyzed the dynamics inside of telco clouds.
We identified 3 main ELEMENTS of Telco Cloud:
Workload – INPUT, operators have NO INFLUENCE
MOBILITY – how user move
Objectives – requirements, specification
Setup – parameters that can be modified, configuration

17. Multi-Model OUR MAIN CONTRIBUTION is multi-model
BY MULTI-MODEL I MEAN USING EXISTING models of different parts of Telco Cloud: models for DATA CENTERS, NETWORK DELAYS, MOBILITY and REQUEST GENERATION and COMBINING them to simulate the WHOLE SYSTEM

18. Multi-Model Parameters
Objectives
Performance Overhead (e.g., migration cost)
NOW I will present in details SELECTED PARAMETERS we used for modeling and I will FOCUS on the parameters that have INFLUENCE on the PERFORMANCE OVERHEAD of DISPERSED COMPUTATIONS (e.g., costs of GEOGRAPHICAL migrations)

19. Multi-Model Parameters
Stateful Applications
Objectives
Workload
User’s State
Response
State
1010
User’s State Size
Total Size of
User’s Requests
Request
When modeling WORKLOAD we include a parameter describing USER’S STATE
# In case of STATEFUL APPLICATIONS
RESULTS of processing # a REQUEST is not only a # RESPONSE but also # USER’S STATE
Therefore a RESPONSE doesn’t only depend on the request but also on the user’s state
IN OUR MODEL User’s State Size IS PROPORTIONAL to the total size of User’s Requests: More requests -> Bigger User’s State

20. Multi-Model Parameters
Objectives
Workload
User’s State
Mobility
User’s State has to follow User’s movements
State
Next parameter of workload is USER’S MOBILITY that DEFINES how users MOVE around the simulated area,
We use TWO DIMENSIONAL, MULTI MODAL MOBILITY MODEL with ON-AVERAGE UNIFORM DISTRIBUTION OF USERS
When Mobile Users use Stateful Applications # User’s State has to follow User’s movements and when user crosses the boarder between areas covered by two different Proximal Data Centers # he changes PDC, and his requests will be processed by new PDC then # USER’S STATE MIGRATION is necessary
User changes PDC

21. Multi-Model Parameters
Objectives
Workload
User’s State
Mobility
Setup
Data Center Catchment :
Data Center Catchment – # Rato between the NUMBER of Radio Base Stations and the NUMBER of Data Center
HOW MANY RADIO BASE STATIONS EACH DATA CENTER HAS TO HANDLE
Ratio between RBS and DC

22. Multi-Model Parameters
Objectives
Workload
User’s State
Mobility
Setup
Data Center Catchment
Virtual Machine Life-Cycle
To CAPTURE the performance OVERHEAD of migrations we include a model of VIRTUAL MACHINE LIFE-CYCLE

23. Virtual Machine Life-Cycle
AT THE BEGINNING all VMs - INACTIVE state. WHEN 1st request arrives to a DC - VM is initiated
process requests OR migrations at the same time (migrations higher priority) | no requests nor migrations -> IDLE # VM is terminated if IDLE state lasts for longer than tidle seconds # VM termination takes tterm seconds

24. Simulation Showcase
ANOTHER CONTRIBUTION is we IMPLEMENTED a PROTOTYPE of TELCO CLOUD SIMULATOR
Simulation showcase shows some CAPABILITIES of simulator that we have implemented
NOT SOLVING any optimization problem etc.

25. Simulation Showcase Setup
16 Radio Base Stations (4x4 layout)
Catchment of Data Centers varies
SQUARES – Radio Base Stations CELLS
COLOR RECTANGES – CATCHMENT of Data Centers

26. Results Small Catchment  Few Devices  Inactivity
Bigger Catchment  Longer Stay  More to Migrate
Results
Static
Mobile
(Y) TIME SPENT in each VM STATE #(colors) in the system per DC CATCHMENT #(X axis); CONSTANT number of REQUESTS – constant processing time -> Other states are interesting; 30 DEVICES per CELL; LEFT – Static users # (1:1) INACTIVE (dark blue) few users in DC catchment
# RIGHT – Mobile users; MIGRATION (orange) higher overhead for smaller catchment however is not proportional – 1:1 – 47%, 1:8 26% REASON: Users’ states do not have the time to grow; Simulation showcase presents CAPABILITIES of simulator not intended to NOT SOLVE any optimization problem etc.

27. Summary
TO SUM UP the presentation I will REMIND our contributions

28. Telco Cloud Dynamics
Workload
Objectives
Setup
We IDENTIFIED fundamental telco cloud DYNAMICS in terms of the RELATIONS between system INPUT, CONFIGURATION and OUTPUT which we call WORKLOAD, SETUP and OBJECTIVES.

29. Multi-Model
We PORPOSED a MULTI MODEL that combines telecommunication and cloud computing INFRASTRUCTURES to simulate the INFLUENCE of users MOBILITY and infrastructure SETUP on the system PERFORMANCE and COSTS.

30. Who can benefit? Telecommunication Operators and Equipment Developers
Model existing infrastructure and plan changes
Researchers
Testing algorithms for resource management
Mobile Applications Developers
Testing behavior of apps in Telco Cloud environment
Our model and simulator can be beneficial for:
TELECOMMUNICATION OPERATORS – to model EXISTING infrastructures and plan FUTURE changes
RESEARCHERS – to TEST ALGORITHMS for resource management
MOBILE APPLICATIONS DEVELOPERS to test behaviour of applications in Telco Cloud ENVIRONMENT

32. Limitations Homogeneous requests (at application level)
All requests contribute to User’s State
Simplified mobile access network model
Tangent, non-overlapping, square cells
No physical layer, channel provisioning and cell load balancing

33. Simulation Showcase Setup
16 Radio Base Stations (4x4 layout)
Catchment of Data Centers varies
Constant capacity of Data Centers
#VMs scales with the catchment
workload is balanced between VMs
480 Mobile Devices

34. Simulation requirements
Scale (number of)
Mobile Devices – hundreds/thousands
Radio Base Stations – tens
Data Centers – several
Length and granularity of simulation (time)
Minutes/hours for mobility patterns
Milliseconds for latency

35. Why simulation? No existing Telco Cloud infrastructure
Extremely high costs of building infrastructure for testing purposes
Huge number of possible setups

36. When to Offload?
Offloading is beneficial when large amounts of computation C are needed with relatively small amounts of communication D
Kumar, K.; Yung-Hsiang Lu, "Cloud Computing for Mobile Users: Can Offloading Computation Save Energy?," Computer , vol.43, no.4, pp.51,56, April 2010

37. Multi-Model OUR MAIN CONTRIBUTION is multi-model
WHAT IS MULTI-MODEL– we combine existing models of different parts
Details of parameters used for modelling

38. Multi-Model Parameters
Workload
Request Generation
1010
HOW REQUESTS ARE GENERATED BY MOBILE DEVICES
Number of requests per session
Size of requests
Inter-request time

39. Multi-Model Parameters
Workload
Request Generation
Resource Requirements
1010
HOW REQUESTS ARE PROCESSED BY DATA CENTERS
CPU cycles used by service
Size of memory and storage used by service
Size of user’s state produced per request

40. Multi-Model Parameters
Workload
Request Generation
Resource Requirements
Mobility
HOW MOBILE DEVICES MOVE
Number of Mobile Devices
Movements of Mobile Devices

41. Multi-Model Parameters
Workload
Setup
Mobile Network
HOW MOBILE NETWORK LOOKS
Number of Radio Base Stations
Dimensions of an RBS cell

42. Multi-Model Parameters
Workload
Setup
Mobile Network
Internet
WHAT DELEY INTERNET INDUCES
Connecting:
RBS with DC
DCs with each other

43. Multi-Model Parameters
Workload
Setup
Mobile Network
Internet
Data Center
DCs - Host multiple services in VM
Number of data centers
Number of servers in data centers
Number of CPUs per server
CPU speed
Memory capacity
Storage capacity
(combined with resource requirements allows to compute computation time)

44. Multi-Model Parameters
Workload
Setup
Mobile Network
Internet
Data Center
Service Placement ?
Where to place services? Proximal vs Remote? When migrate? ?

45. Multi-Model Parameters
Workload
Setup
Objectives
Quality of Service
HOW TO SPECIFY QUALITY OF SERVICE
Application response time
Application throughput

46. Multi-Model Parameters
Workload
Setup
Objectives
Quality of Service
Costs
Total cost of infrastructure
Increased by computing resource dispersion
Reduced by virtualizing some of telecommunication functions