2. www.helsinki.fi
• Why IoT is an interesting use-case
• Issues with current cloud model and IoT
• Requirements from the use-case
• Proposed IoT-compute cloud
• Stakeholders: Cloud, Fog, IoT
• Envisioning Security
• Envisioning Mobility
• Priority computation using SFC
• Conclusion
• Future Vision
28.11.2016 2
Matemaattis-luonnontieteellinen tiedekunta /
Henkilön nimi / Esityksen nimi
Outline
http://blogs-images.forbes.com/jacobmorgan/files/2014/05/libelium_smart_world_infographic_big.png
3. 3
Data generators are increasing!
Source: National Cable & Telecommunications Association, “Broadband by the numbers,” https://www.ncta.com/broad- band-by-the-numbers,
accessed April 22, 2015
Number of connected sensors
Sensor prices over last 25 years
5. 5
Edge & Fog Cloud Computing
Recent proposals to bring the cloud closer to the
end-users!
6. 6
Edge Cloud Computing
Edge voluntary devices with processing power are part of the cloud
Lopez, P. G., Montresor, A., Epema, D., Iamnitchi, A., Felber, P., & Riviere, E. (2015).
Edge-centric Computing : Vision and Challenges. Acm Ccr, 45(5), 37–42.
7. www.helsinki.fi
Fog Cloud Computing
7
In-network devices are part of the
cloud
Bonomi, F., Milito, R., Zhu, J., & Addepalli, S. (2012). Fog Computing and Its Role in the Internet of
Things. Proceedings of the First Edition of the MCC Workshop on Mobile Cloud Computing, 13–16.
8. www.helsinki.fi
Edge & Fog Cloud Computing:
Problem
8
If not completely, the computation is still dependent on
execution at the central cloud, thus includes high network costs!
9. 9
Edge-Fog Cloud
Edge:
1. Consolidation of voluntary compute devices
2. 1-2 hops away from the sensors
3. High mobility and low compute power
4. Ad-hoc direct connectivity within the layer
10. 10
Fog:
1. Network devices with high compute power
2. Installed by cloud providers as backbone to network
3. 5-6 hops away from the sensors
4. High-speed connectivity within the layer
5. Guaranteed connectivity to the edge
Edge-Fog Cloud
12. 12
Advantages of Edge-Fog
1. Computation is done close to the querying users and
sensors → handle high volume of data
2. Mobility of edge layer (or techniques like VM migration)
can cater to mobile applications such as smart
transport system
3. High distribution of computation → exploits parallelism
4. Hierarchical aggregation of intermediate computations
14. 14
Previous Works:
1. Naive Approach
• Calculate all possible assignments → select one with least network
cost
• Too much computation (n! assignments)
• E.g. assign 10 jobs on 10 devices → 3628800 assignments
2. Branch-and-Bound Hungarian Algorithm
𝑎(𝑖,𝑗)∈𝐴
𝐽𝑐𝑜𝑛𝑛 𝑖, 𝑗 ∗ 𝐷𝑐𝑜𝑛𝑛(𝑓 𝑖 , 𝑓(𝑗))
• Famously known as Quadratic Assignment Problem (QAP)
• NP-hard problem, approximation not guaranteed
Workload Assignment on Edge-Fog
Network-Only Cost (NOC)
15. 15
Least Processing Cost First (LPCF)
1
1
3 4 3
J1
J2
J3
J4
J5
8 8
3 2 4
Device Processing Power (𝐷 𝑝𝑟𝑜𝑐(𝑖))
3 4
2 3
1
Job Size (𝐽𝑠𝑖𝑧𝑒(𝑖))
Workload Assignment on Edge-Fog
16. 16
Our Approach
1. Find the assignment with minimum processing time
𝑖,𝑗∈𝐴
𝐶
𝐽𝑠𝑖𝑧𝑒(𝑖)
𝐷 𝑝𝑟𝑜𝑐(𝑗)
𝑥𝑖𝑗
This is Linear Assignment Problem which can be computed in
𝑂 𝑛3
2. Compute all possible permutations of assignments having
same processing cost (smaller subset)
3. Use assignment with least network cost
Workload Assignment on Edge-Fog
19. 19
Several further issues
1. Network cost while transporting data from central
database?
2. Guaranteeing stable system while considering attacks on
volunteering Edge resources?
3. Incentive mechanisms for Edge resource owners?
4. Light weight VM design for Edge devices to support
computations
5. Providing network functions in Edge-Fog cloud?
6. Ideal network model for Edge-Fog cloud? ICN? MPTCP?
TCP?
And so on…