Beyond Boundaries: Leveraging No-Code Solutions for Industry Innovation
Design and emulation tools for serverless edge computing
1. An Architectural Framework for
Serverless Edge Computing:
Design and Emulation Tools
Claudio Cicconetti
Marco Conti
Andrea Passarella
2. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What do we want
from edge computing?
• Actually, we just want it to be fast, as in “faster than the
cloud”
3. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What are the barriers against
edge computing being fast?
1. Latency to set up VM/container/run-time environment
2. Edge application migration times
4. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What can we do about it?
1. Remove set up procedures
2. Remove migration
5. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Serverless computing
6. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Serverless computing
clients
pool of
invokers
OpenWhisk
controller
CockroachDB
NGINX
Kafka
(1)
(6)
(4)
(5)
(7)
(2)
(3)
7. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Serverless edge computing
edge
computing
domain
A
UT
IN
OUT
B
D
E
F
C
8. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What do we need for that?
• All edge domain entry points must know where the
lambda functions are
• The position of lambda functions does not change too
frequently
• A decision on where to execute a lambda function must
be taken super fast
• Clients don’t care of past executions
9. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What do we need for that?
• Let’s think about this…
10. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
What do we need for that?
• Wait a minute, it’s like the Internet Protocol!
11. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Same problem,
same solution
• Computation devices act as destination hosts
• Entry points act as routers
• They use forwarding tables to direct traffic
• These tables are kept up-to-date by means of routing
protocol operating in background
12. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
All that glitters is not gold
• There is an intermediate time scale between lambda
execution and lambda configuration!
A. Time-varying load conditions
B. Time-varying network conditions
13. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Example
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 500 1000 1500 2000 2500PMF
Latency (ms)
Left
Right
14. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Example
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 500 1000 1500 2000 2500PMF
Latency (ms)
Left
Right
15. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Example
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 500 1000 1500 2000 2500PMF
Latency (ms)
Left
Right
16. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Research challenge
• How to update forwarding table weights to follow
A. Computation load
B. Network load
17. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Proposal
• Set weight = average latency*
(after all that’s what really matters, remember?)
• Use one of the following forwarding algorithms
• Draw destination randomly, probability = 1/weight
• Short-listed weighted round-robin**, quantum = weight
• Select destination with smallest weight = smallest delay
18. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
How do we know it works?
• Well, of course we implemented an emulator and ran a
zillion experiments
PU#1PU#11 2
3
waiting scheduled memory
1
2
3
PU#1PU#24 5
4
5
container
queued
requests ...
e-table
weight
calculator
1
2
3
4
5
6
8
7
9
client e-computer
handlers
19. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
A simple experiment
• One edge router
• Nine devices offering lambda functions
• Chain topology, to emphasise different network delays
Edge router
Edge clients
Edge application servers
…
Background traffic
20. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
A simple experiment
0
0.2
0.4
0.6
0.8
1
2 3 4 5 6 7 8 9 10
Averageloadpere-computer
#e-computer
WRR no congestion
WRR congestion
Random no congestion
Random congestion
21. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
A simple experiment
0
50
100
150
200
250
300
350
400
0 20 40 60 80 100 120 140 160 180
95thpercentileoflatency(ms)
Load (number of terminals)
WRR no congestion
WRR congestion
Random no congestion
Random congestion
22. C. Cicconetti, M. Conti, A. Passarella - Presenter: @ccicconetti
An Architectural Framework for Serverless Edge Computing: Design and Emulation Tools - CloudCom 2018
Takeaways
• Edge computing
• Not the new cloud, but rather a faster cloud
• To make it a reality, it must be
• Low-latency
• Reactive
• Easy to use
• Serverless computing is a possible starting point