Edge computing system for
large-scale distributed
sensing systems
Miloš Simić, Milan Stojkov, Goran Sladić, Branko Milosavljević
Faculty of technical sciences, University of Novi Sad
Plan
• Motivation
• Main problem
• How to solve a problem
• Recap
• Future work
• Questions
Motivation [1/2]
• Nowadays we are facing a massive shift away from the standard,
centralized computing model that is provided through cloud computing
paradigm.
• The shift is now towards the distribution of computing power back to the
edge of the network.
• Edge network is the idea of connecting sensors to programmable
automation controllers (PAC) which handle processing, storage, and
communication.
• The basic concept of edge computing is to leverage new generation
technologies, processes, services, and applications that are built to take
an advantage of new infrastructure.
Motivation [2/2]
• Extending Cloud computing paradigm
• Try to overcome idea “one size fits all”
• Help Cloud in new use cases and applications
like autonomous cars, delivery drones, sensing
systems etc.
Main Problem [1/2]
• The problem with the current model is that heavily
involves centralized architecture.
• Issues such as latency or small-time delays, security,
privacy, network reliability, performance and many
others are extremely difficult to completely overcome in
centralized computing models.
• Even a small problem can set a motion to bigger
complicated issues.
Main Problem [2/2]
• Cloud computing has driven an incredible
amount of innovation over the last 20 years.
• It is too far away from the most interesting things
that are happening around us.
• It is too slow to react to change in the real world.
• Edge computing is here to help Cloud computing
How to solve a problem [1/4]
• One possible solution is Edge computing
• Edge, is right next to where the action is.
• It's already in the densest urban areas where most of
humans and their devices live, and where commerce
happens.
• Put processing closer to the edge of the network pre-
process data and send to the cloud
How to solve a problem [2/4]
• System we propose is influenced by few proven
systems in both academia and industry:
• Orchestration engine from Google called Borg, that
helps Google run their workloads efficiently with
maximum resource utilization
• Chord, lookup protocol by MIT for fast data lookup in
distributed p2p environment
• Log-structured merge-tree (LSM tree) append only
key-value data structure for efficient data storage
presented by Patrick O'Neil.
How to solve a problem [2/4]
• System we propose is influenced by few proven
systems in both academia and industry:
• Orchestration engine from Google called Borg, that
helps Google run their workloads efficiently with
maximum resource utilization
• Chord, lookup protocol by MIT for fast data lookup in
distributed p2p environment
• Log-structured merge-tree (LSM tree) append only
key-value data structure for efficient data storage
presented by Patrick O'Neil.
How to solve a problem [2/4]
• System we propose is influenced by few proven
systems in both academia and industry:
• Orchestration engine from Google called Borg, that
helps Google run their workloads efficiently with
maximum resource utilization
• Chord, lookup protocol by MIT for fast data lookup in
distributed p2p environment
• Log-structured merge-tree (LSM tree) append only
key-value data structure for efficient data storage
presented by Patrick O'Neil.
How to solve a problem [3/4]
• Separate geographic area where sensing systems are,
in 'micro data-centers' that collect and (pre)process data
• Every node in micro datacenter should run two
processes:
• minion process that communicate with the cloud,
container engine process, and rest of the nodes in
micro datacenter (replication, lookup)
• container engine that runs applications in containers
providing isolation and maximum resource utilization
How to solve a problem [4/4]
• We propose few programming models for users of this system:
• batch jobs, for standard batch processing over some collection of data
• events, this type of job should react when some specific event happened
in the system, or pass some predefined threshold
• streaming jobs, should continually do some (pre)processing on data as it
gets (long running jobs)
• services, user defined jobs that do not fit in any of the previous categories
• Jobs contains DAG when transits from operation to operation to prevent data
loss and starting jobs from start if fail happened
Overall architecture
Recap
• Cloud is great tool, but one size does not fit all
• Edge computing can help in (pre)processing
information's and speed up time to market
• Separate region in micro data centers that store
and (pre)process data at the edge of the network
• Offer users programming models that they are
familiar with from cloud applications
Future work
• Monitoring of such a system
• Configuration and secrets dissemination
• Security
• Service mashes
Questions?
Thank you for your attention
‘Containers are great, let’s run them everywhere’

Edge computing system for large scale distributed sensing systems

  • 1.
    Edge computing systemfor large-scale distributed sensing systems Miloš Simić, Milan Stojkov, Goran Sladić, Branko Milosavljević Faculty of technical sciences, University of Novi Sad
  • 2.
    Plan • Motivation • Mainproblem • How to solve a problem • Recap • Future work • Questions
  • 3.
    Motivation [1/2] • Nowadayswe are facing a massive shift away from the standard, centralized computing model that is provided through cloud computing paradigm. • The shift is now towards the distribution of computing power back to the edge of the network. • Edge network is the idea of connecting sensors to programmable automation controllers (PAC) which handle processing, storage, and communication. • The basic concept of edge computing is to leverage new generation technologies, processes, services, and applications that are built to take an advantage of new infrastructure.
  • 4.
    Motivation [2/2] • ExtendingCloud computing paradigm • Try to overcome idea “one size fits all” • Help Cloud in new use cases and applications like autonomous cars, delivery drones, sensing systems etc.
  • 5.
    Main Problem [1/2] •The problem with the current model is that heavily involves centralized architecture. • Issues such as latency or small-time delays, security, privacy, network reliability, performance and many others are extremely difficult to completely overcome in centralized computing models. • Even a small problem can set a motion to bigger complicated issues.
  • 6.
    Main Problem [2/2] •Cloud computing has driven an incredible amount of innovation over the last 20 years. • It is too far away from the most interesting things that are happening around us. • It is too slow to react to change in the real world. • Edge computing is here to help Cloud computing
  • 7.
    How to solvea problem [1/4] • One possible solution is Edge computing • Edge, is right next to where the action is. • It's already in the densest urban areas where most of humans and their devices live, and where commerce happens. • Put processing closer to the edge of the network pre- process data and send to the cloud
  • 8.
    How to solvea problem [2/4] • System we propose is influenced by few proven systems in both academia and industry: • Orchestration engine from Google called Borg, that helps Google run their workloads efficiently with maximum resource utilization • Chord, lookup protocol by MIT for fast data lookup in distributed p2p environment • Log-structured merge-tree (LSM tree) append only key-value data structure for efficient data storage presented by Patrick O'Neil.
  • 9.
    How to solvea problem [2/4] • System we propose is influenced by few proven systems in both academia and industry: • Orchestration engine from Google called Borg, that helps Google run their workloads efficiently with maximum resource utilization • Chord, lookup protocol by MIT for fast data lookup in distributed p2p environment • Log-structured merge-tree (LSM tree) append only key-value data structure for efficient data storage presented by Patrick O'Neil.
  • 10.
    How to solvea problem [2/4] • System we propose is influenced by few proven systems in both academia and industry: • Orchestration engine from Google called Borg, that helps Google run their workloads efficiently with maximum resource utilization • Chord, lookup protocol by MIT for fast data lookup in distributed p2p environment • Log-structured merge-tree (LSM tree) append only key-value data structure for efficient data storage presented by Patrick O'Neil.
  • 11.
    How to solvea problem [3/4] • Separate geographic area where sensing systems are, in 'micro data-centers' that collect and (pre)process data • Every node in micro datacenter should run two processes: • minion process that communicate with the cloud, container engine process, and rest of the nodes in micro datacenter (replication, lookup) • container engine that runs applications in containers providing isolation and maximum resource utilization
  • 12.
    How to solvea problem [4/4] • We propose few programming models for users of this system: • batch jobs, for standard batch processing over some collection of data • events, this type of job should react when some specific event happened in the system, or pass some predefined threshold • streaming jobs, should continually do some (pre)processing on data as it gets (long running jobs) • services, user defined jobs that do not fit in any of the previous categories • Jobs contains DAG when transits from operation to operation to prevent data loss and starting jobs from start if fail happened
  • 13.
  • 14.
    Recap • Cloud isgreat tool, but one size does not fit all • Edge computing can help in (pre)processing information's and speed up time to market • Separate region in micro data centers that store and (pre)process data at the edge of the network • Offer users programming models that they are familiar with from cloud applications
  • 15.
    Future work • Monitoringof such a system • Configuration and secrets dissemination • Security • Service mashes
  • 16.
    Questions? Thank you foryour attention ‘Containers are great, let’s run them everywhere’