Paper Discussion 11b: The Swarm at the Edge of the Cloud

[anguyen0204]

Summary:
@albero94: Discusses about the implication of swarmOS into personal devices and its potential effects on people as a whole and the research team's progress on it. His thoughts are apprehensive about the project overall
@ratnadeepb: Talks about how real time and design time can be fluid together and his feelings overall towards the paper are disturbing
@gkahl: Feels that the paper isn't fleshed out enough
@nikorev: "Were there any real-world test runs of these swarmlets conducted? They discuss possible applications of SwarmOS, but has this been extended into another paper with real-world testing?"
@s-hanna15: "If there are two nodes providing conflicting information, how does it choose which one to believe?"
@huachuan: Talks about the hierarchy of the swarm and how effect can swarmOS be inetgrated
@rebeccc: "I don't understand this paper's definition of Adaptive as it pertains to their system. What does it mean for the distinction between "design-time" and "run-time" to be blurred?"
@pcodes: feels that the paper is pretty vague and brings up a good point how edge computing can be put into all of this
@bushidocodes: feels that the entire paper is shoddy and that there needs to be more evaluation and work done and overall it makes him uncomfortable
@chandaweia: saw that the paper provides too many vague buzzwords and does not flesh out the entire premise enough and that they are not providing with enough information.

Shared Consensus:

1. This paper does not bode well and needs to be more fleshed out since it is too vague
2. There are too many assumptions and how can this new swarmOS be efficiently integrated into everything?

[albero94]

Reviewer: Alvaro Albero
Review Type: Analytical

Problem Domain

There is an increasing number of embedded devices that are connecting the cyber and the physical world. But so far these devices have been working as independent components instead of seamlessly integration.

Main Contributions

The authors propose a platform to wrap these devices and provide a way to deploy and operate them, they use the term tera-swarm. They use as a reference the mobile platform and they want to achieve a similar system to bring millions of developers to this platform where they can create and deploy applications, called swarmlets, that will recruit resources such as sensors or actuators from the swarm.

Questions

• In Section C4, they talked about applications being dynamically composed and recomposed and they mention that optimizations cannot be done at design time, must be done at run time. What do they mean with this? Why is it different here than in other systems?

• How will the swarmOS be able to interface with all the types of embedded devices and cloud providers in order to have a unique platform? Are they expecting everyone to accept and migrate to their solution?

• This is a paper from 2014 where they are providing their vision. What has been the progress since then? Has this project kept evolving? Is work still being done in this direction or workshops like the one in 2013 still taking place?

Critiques

• They make a comparison to the mobile ecosystem and say that they want to create a similar platform. I think that the way iOS and Android became the dominant platforms in the mobile world was not through decided or organized top down. Developers and consumers naturally moved towards those platforms because they were offering a better experience. This paper looks more like a top down organization where they explain that we should all move towards this one platform that does not exist yet.

• They talk about this swarm where all embedded devices are interconnected and a swarmOS that controls the services and resources provided to swarmlets. This looks dangerous to me, you have this unique network that has access to all the imaginable sensors, actuators, etc. and you expose them through an OS to all the applications (swarmlet) that anyone can build. I feel this could be a huge platform with huge potential to do good things, but also bad things, and you are giving all this power to random developers and organizations without knowing what they will use it for.

[ratnadeepb]

Reviewer: Ratnadeep Bhattacharya
Review Type: Comprehension

Problem Domain

The boundary between the cyber and the physical world is becoming hazier due to the sheer number of sensors and actuators embedded in the physical world. The authors argue that this presents an opportunity to provide mega-scale services that involve multiple vendors and dynamic use of resources and available data.

Main Contributions

The authors list out the challenges towards this goal and make a strong case for a "SwarmOS", an all encompassing framework that includes the cloud backbone, the various sensors and actuators, all available data and data models. A further case is argued about seemingly uncontrolled access to all these resources by vendors aiming to provide these services. The paper then goes on to list challenges, possible technologies and research directions to solve those challenges and applications that could result.

Questions

In C2, the authors mention that virtualization of computing resources is not against real-time or time-sensitive services but more work is required to manage the quality of service in networks and temporal isolation is required in processors. I am not quite sure what exactly is meant by this. An example might have helped!

Critique

1. I am paranoid as far as the Internet and related privacy violations. As such this paper sends shivers down my spine. The sheer opportunity for privacy and security violations that could be created through the indiscriminate sharing of data with little control is scary. I don't think the paper addresses this subject in any seriousness.
2. The paper is sort of a "vision" paper and as such does not treat any particular topic to any depth. Their references section too is pretty thin and there are not many opportunities to dive down deeper in the context of this paper.

[gkahl]

Contributor: Greg Kahl

Review Type: Comprehensive

Problem

There are more and more edge sensors and embedded systems being implemented in order to improve quality of life and collect data. However, these sensors and small systems are not all connected for an efficient use of the vast amount of data they are collecting.

Contribution

This paper suggest their TerraSwarm system to connect and manage all of these small devices on the system. They believe it will be a platform for future development to help bring all of the devices together for a better use and processing of the huge amount of data these edge sensors collect.

Questions

1 - They claimed that data doesn't need to be communicated or stored if they can predict it with models. What does this mean? How is the data useful if you are never actually using it and just trying to predict it?
2 - They discuss the use of the platform for autonomous vehicles, but I felt like they didn't really explain scheduling and critical systems such as cars on such a large system. How would they handle that to ensure the safety of the users?
3 - As we have read about in many other papers, security on the edge devices is very important. They briefly brush this off by saying they will develop energy efficient hardware support for security features what does this mean?

[nikorev]

Reviewer: Niko Reveliotis
Review Type: Comprehension

Problem Being Solved

With the rise of IoT devices, there has been a "swarm" of sensors and actuators being introduced to these devices. The paper describes these sensors and actuators giving the cloud "eyes, ears, hands, and feet". This paper introduces "swarmlets" which will try to guarantee secure and proper operation of these sensors through an open and universal platform.

Main Contributions

• TerraSwarm/"swarmlet" application tools/framework, named SwarmOS, which provide the ability to "dynamically recruit resources such as sensors, communication networks, computation, and information from the cloud; to aggregate and use that information to make or aid decisions; and then to dynamically recruit actuation resources — mediating their response by policy, security, and privacy concerns."

Three Questions

1. How would SwarmOS be able to get the market-share of both sensors and the cloud to become a player in the game? How do they expect to get everyone on board?
2. What tools are they referring to in needing to blur the distinction between "design time" and "run time"?
3. Were there any real-world test runs of these swarmlets conducted? They discuss possible applications of SwarmOS, but has this been extended into another paper with real-world testing?

[s-hanna15]

Reviewer: Sam Hanna
Review Type: Critical

Problem Being Solved:
This paper talked about what could be accomplished if all of the IoT sensors and devices could be combined into the TerraSwarm. Currently, all IoT devices are disconnected and doing different projects for different purposes. They ask the questions of what could be accomplished if all IoT devices could be utilized to solve problems.

Important Areas:
This paper focuses on the idea of a TerraSwarm. This would focus on utilizing swarms and swarmlets in order to create more of a smart-city environment. They would create and utilize SwarmOS, a stable base operating system, in order to have access to all nodes and optimize the use of the swarms.

Questions:

1. If there are two nodes providing conflicting information, how does it choose which one to believe?
2. They mention briefly security being built in from the start, what steps are they taking to ensure that this can’t get hacked or a malicious application (swarmlet) does not purposely join the system?
3. What is the incentive for IoT device makers to join this? They use phone applications as a reference, but phones already had a huge use outside of the applications. How would this track with no strong user-base prior to getting developers?

Critiques:

1. They are vague on the details of a lot of the technical parts of how they are going to deal with large numbers of data, with nodes leaving the system, and with privacy concerns.
2. They talk very briefly about privacy and security, but really not enough considering the HUGE privacy implications of connecting the number of sensors they are talking about.
3. They talk about a lot of theoretical uses for this but do not have any real tests of the system or the different parts of the system.

[huachuan]

Reviewer: Huachuan Wang
Review Type: Comprehension

Overview

This paper introduced the swarm which is an extension from the cloud to the physical world and enabling services that are directly embedded in the physical world. This paper also proposes swarmlets which is the adoption of an open and universal platform to enable a multiplicity of distributed sense and control applications.

Contribution

This paper proposed the swarmlets which are characterized by their ability to
(1) dynamically recruit resources such as sensors, communication networks, computation, and information from the cloud
(2) to aggregate and use that information to make or aid decisions
(3) and then to dynamically recruit actuation resources mediating their response by policy, security, and privacy concerns.
This paper also addressed the swarmlets's wide range of challenges and opportunities. This paper suggested a collaborative environment t to generate a broad range of innovative ideas and solutions.

Questions

1. I am confused about the dynamic hierarchical graph of TerraSwarm, a node can aggregate services to define new service, but could a node be part of several services at the higher level? Also, the author describes that the edge has many representations, does that mean each edge has all the representations?

2. The author states that in order to fully achieve the potential of the swarm, they need a SwarmOS. However, given many existing operating systems already been successful, can we directly use or make modifications with these operating systems to implement the swarmlets? What are the special requirements that the existing operating system could not meet?

[rebeccc]

Reviewer: Becky Shanley
Review Type: Comprehension

Problem being addressed

The paper identifies an unexplored potential in using the "sensory swarm" as one uniform data aggregate of the physical world that could improve social problems and quality of life, called the TerraSwarm. They identify this as a useful field and then address the biggest problem standing between this reality: a lack of universal architecture/communication protocols that allow developers to create something useful on a large scale.

Main Contributions

This paper contributes to solving the problem of a lack of uniform development environment and architecture for the TerraSwarm by proposing a SwarmOS, which they compare to the idea of Android in that it is specific to IoT devices and their sensors.

Questions

1. I don't understand this paper's definition of Adaptive as it pertains to their system. What does it mean for the distinction between "design-time" and "run-time" to be blurred?
2. In section C3 it mentions that "Security-related technologies and techniques such as static analysis, hazard analysis, and elliptic curve cryptography will also prove useful" but didn't expand on what any of these will be useful in achieving other than a broad "security".

[pcodes]

Reviewer: Pat Cody
Review Type: Comprehensive

Problem Being Solved:

As the number of sensors in the IoT explodes over the coming years, it will be important to find effective ways to utilize them. Many sensors currently only serve a single service or function, when there is much more power in re-using them and combining sensors for multiple services.

Main Paper Contributions

This paper establishes several abstractions, including "swarms", or groups of IoT sensors, and the "terraswarm", a global cyber-physical network (comprised of swarms and applications). By allowing multiple applications to utilize the same sensors, the collected data can be better used as less of it will be wasted. This also allows for more complete applications/services, as their idea of "SwarmOS" would give app developers a framework to build apps that utilize everything in the Terraswarm.

Questions

• How do the paper authors envision maintaining privacy with this system? It's a pretty vague paper in all aspects, so I'm not surprised they didn't address this, but merging smart homes and smart cities into the same network means that there will be a lot of personal data available. How would app developers safely be able to access this? How would they prevent super sophisticated ad delivery (would they?)
• Can we adapt other existing IoT abstractions to work with the swarm concept? For example, how does edge computing fit into all of this?
• How are we moving away from "algorithms" and towards "dynamics", "distribution", etc.? Aren't these still algorithms?

[bushidocodes]

Reviewer: Sean McBride

Review Type: Critical Review

Problem Being Solved:

This paper is Ted Nelson Xanadu levels of crazy, so it's hard to trace a coherent argument or problem. It is disturbing similar to the Radiohead song Fitter Happier.

I suppose the question is sort of meta or philosophical about how sensors and IOT devices should be organized given that:

• sensors around to proliferate to 1000x per person
• data is more valuable when accumulated

Main Contributions:

1. Tries to envision a very decentralized bazaar-not-Cathedral bottom-up approach to IOT, whereby arbitrary sensors, actuators, and edge resources share some common communications and control mechanisms ("SwarmOS") and can be composed together into graphs of services called "swarmlets." This is envisioned to be connected to seemingly everything across the world, creating a global network called the "TerraSwarm."
2. Provides sci-fi examples of "TerraSwarm Apps" such as smart jukeboxes that dynamically generate soundscapes based on the preferences of customers, but also have some sort of Orwellian knowledge of where the Doctors are, so it can bark orders at them when there are emergencies.
3. Gives me nightmares.

Questions:

1. How was this a peer-reviewed article that led to funding of a research institute and major corporations? Is one of the authors an eccentric genius or something?
2. Who own these sensors that can be arbitrarily enlisted? What mechanism actually manages scarcity among these competing TerraSwarm Apps?
3. What does it mean to have post-algorithmic computation?

Criticisms:

1. As mentioned above, this feels more Ted Nelson Xanadu than Tim Berners-Lee. Ideas and political philosophy just kinda swirled together without justification. The use of buzzwords and adjectives reminds me of the book House of Leaves, where there is nonsensical narrative with a hidden message that gives off a vague sense of dread. I haven't been able to discern the hidden message in this article. The paper mentions "battlefield management systems," "soundscapes," "control as a service," "Total Awareness," "swarm-in-the-loop simulation"
2. Different and contradictory statements are thrown out without much thought. Everything is decentralized, but unified. Everything is dynamic and arbitrarily composed, yet resilient to collapse. Things are dynamic and impromptu, yet somehow respect the time limitations of real-type systems. etc. etc. etc.
3. Figure 2 looks like something a mad scientist would draw.
4. The references mostly seem to be the authors' own works

[rachellkm]

Reviewer: Rachell Kim
Review Type: Critical

Problem Being Solved:

The increase in sensors and actuators in embedded devices, called the sensory swarm, poses concerns for deployment, operation, and access control. The author proposes a universal platform to mitigate these technical and non-technical challenges in swarm devices and their applications.

Main Contributions:

The authors propose an open, universal platform that swarm developers may leverage to develop applications (“Swarmlets”). They cite the need for an architecture (“SwarmOS”) that would efficiently manage the resources of a multitude of heterogeneous sensors and actuators. Moreover, they outline several possible approaches and key areas of research that could potentially contribute to the manifestation of the TerraSwarm system.

Questions:

1. The paper mentions that conventional methods of using cloud services to store data would not be enough to support the needs of TerraSwarm devices and applications and instead suggests using simulation models as a different, more scalable approach. I’m not sure if I understood this correctly, but is this implying that the system (specifically, the swarmlets) will rely heavily on mocked data simulated by swarm devices?
2. They mention that swarm technology would have “unprecedented robustness in the face of infrastructure collapse,” but how or on what basis/idea are they making this claim? I personally thought it would be the opposite.

Critique:

1. Perhaps because of the fact that we had just read a paper about the issues in trusting the physics in sensor data, I felt that the security issue was a much larger and serious problem than this paper took time to address, considering that they are proposing a system to handle a network of trillions of sensor devices.
2. This paper describes what needs to be done and what the TerraSwarm should be, but not much of the state of current research and efforts that have already been accomplished vs. what is lacking. I think that would have been nice information to include if it were possible.

[chandaweia]

Reviewer: Cuidi Wei
Review Type: Critique
Problem being solved
Most of sensor-based systems are targeting a single application or function. There is little technology to integrate the cyber and physical world. Therefore, sensory swarms offers an unprecedented ability to monitor and act on a range of evolving physical quantities.

Main contributions
This paper proposes the TerraSwarm, which dynamically recruits resources and information from the cloud, aggregates and use information to make decisions, and dynamically recruits actuation resources to mediate their response by policy, security, and privacy concerns.

Questions and Critiques

1. This paper gives little information how the TerraSwarm improve accuracy and security. We don’t know how the TerraSwarm achieves the security and provides the limited source for Swarmlets.
2. If a smarter approach can build and refine models of the data sources, how to define normal behavior of the data source? What’s the accuracy of checking the normal behavior? Are there any inaccurate results? If yes, how to prevent the inaccurate results?
3. I think this paper talks about many vague words, for example, the system theoretic formulation addresses privacy concerns by defining filters that release useful information. However, we know it difficult to filter the exact useful information. How to achieve this in this paper? The paper provides little results of testing the TerraSwarm.

[gparmer]

@anguyen0204 Where is the summary of everyone's summaries?

[anguyen0204]

@anguyen0204 Where is the summary of everyone's summaries?

updated