Mobile edge computing-based distributed architecture in mobile crowdsensing
Hanane Lamaazi23:152021
One central cloud is too slow, too costly and blind to Bluetooth devices
Smartphones and wearables collect data about the world around them, sometimes without their owners knowing, and that data is typically shipped to a single central platform that struggles to answer in real time and can miss devices sharing over Bluetooth. Khalifa University postdoctoral researcher Hanane Lamaazi studies IoT, edge computing and crowdsensing. Can servers placed near users cut the delays and recruit better workers on a fixed budget?
See how a distributed edge architecture selects crowdsensing workers by service quality and budget
Hanane Lamaazi, a postdoctoral fellow at Khalifa University, explains how mobile crowdsensing collects data through people's smartphones and wearables, and why sending everything to a centralized platform causes delays, missed devices, computing complexity and high cost. She argues that a distributed architecture built on mobile edge computing overcomes these limits, and presents her team's solution that selects crowdsensing workers by quality of service under a budget.
“So do you know that your personal information can be useful for a third party and for like, for example, some companies and some applications? And do you know where and how it is stored?”
“The data is sensed, collected, and shared automatically and sometimes even without the knowledge of the user.”
What you will learn
Distinguish participatory from opportunistic crowdsensing by who controls the sensing, how reliable it is and what it costs.
Identify the four main limitations of centralized crowdsensing platforms: delay, undiscovered non-connected devices, computing complexity and cost.
Explain how an edge layer between mobile devices and the cloud offloads computation, lowers latency and reduces the risk of losing all data at once.
Describe how local edge nodes can select workers by quality of service, including device characteristics and reputation, within a budget.
Recognize what makes an IoT environment heterogeneous in devices, transmission technologies and data.
Key takeaways from the speaker
Edge Computing Technology, Distributed Architecture, Worker Selecetion in Mobile Crowdsensing
Details
Level
Intermediate
Format
Talk
Technical depth
Introductory technical
For
Student, Early career, Mid-career
Length
23:15
Published
June 2021
Questions this talk answers
What is the difference between participatory and opportunistic crowdsensing?
In participatory crowdsensing users volunteer and decide what, when and where to sample, and are often paid, which makes it more expensive. In opportunistic crowdsensing data is sensed and shared automatically, sometimes without the user's knowledge, and depends on the phone's condition.
Why do centralized mobile crowdsensing platforms struggle?
They are slow to answer real-time requests because of the volume of heterogeneous data, cannot discover devices sharing over Bluetooth or Wi-Fi Direct, are complex to compute on, and are expensive because they need large storage and processing capacity.
What role does edge computing play between phones and the cloud?
Edge servers sit near users as an intermediate layer that filters, aggregates, processes and stores data, answering small requests directly and forwarding larger ones to the cloud.
How does a distributed architecture reduce the risk of losing data?
Because data is spread across several servers, an attack or crash does not wipe out everything, and missing information can be estimated from what the other servers still hold.
How does the proposed solution choose crowdsensing workers?
Each local edge node in a cluster selects the best workers by quality of service, covering device characteristics and worker reputation, and keeps recruiting until the budget is reached.
About this talk
Deploying Mobile Edge Computing (MEC) in Mobile Crowdsensing (MCS) can help address many challenges the centralized MCS solutions are facing such as delays in answering real-time requirements due to the centralized nature of the solution, discovering and selecting non-connected devices in the Area of Interest (AoI), and dealing with the complexity of data computation. In this talk, we will discover what is edge computing technology? why it is used and how it can overcome the limitation of centralized platforms? and How it can help to improve the worker selection/recruitment in MCS?
Participatory vs opportunistic crowdsensing (66 s)
What edge computing actually does (81 s)
Offloading work from phones and the cloud (77 s)
Speaker: Before starting, I had some specific questions. So do you know that your personal information can be useful for a third party and for like, for example, some companies and some applications? And do you know where and how it is stored? So this is what will be specifically the talk of today. So I will cover the use of mobile edge computer and distributed architecture and deployed in a mobile crowd sensing environment. So my name is Hanae Lamaazi. I'm a postdoctoral law at the Center for Cyber Physical System at Khalifa University in UAE. So my talks will be generally about introducing the concept of Ethernet of things and crowd sensing, then highlighting the difference between centralized and distributed architecture, introducing the concept of edge computer and concluding by one of our recent research and the recent proposed solution that we call
OFSEC. So as we know that Internet of things, it's an emerging technology that try to connect everyday objects to the Internet. So it covers all domains of our lives, including healthcare, home automation, transportation, cybersecurity, and even industry. So here, as we can distinguish that by the IoT analytics that is done in 2020, that 1,414 projects deployed to Internet of Things concepts, where manufacturing and industrial are the most common ones with 22%, followed by transportation, and then by energy. So here, just to look to see any impact of including the Internet of things in different domains and how they are competing to use this emerging technology. So, as we say, Internet of things, it's connecting everyday objects, which means that generating a set of a huge number of the data that it will be processed, filtered, and stored later. So crowd sensing,
it's like it's like a specific case or sub domain of it's enabled technology for Internet of things that has a main objective is to collect the information or to collect the data by deploying a set of users or what we call in crowd, the term used in crowd sensing, it's workers that use mobile devices, generally some smartphones and wearable devices.
And executing some specific tasks while generating or collecting more information about a specific phenomenon. So there is two kinds or two types of crowd sensing. The first one is the participatory crowd sensing and the second one is the opportunistic one. So in participatory crowd sensing, the users volunteer to participate while this is not the case for opportunistic.
The data is sensed, collected, and shared automatically and sometimes even without the knowledge of the user. With participatory, the user decides or determines how and when and what to sample and where to sample or to collect the data. While this is not the case for opportunistic, that gather a large amount of data that will be processed later on, but in a small time.
So it doesn't require too much time. So for participatory, avoid the phone issues like, for example, application crash, bad battery life and so on. So for example, if you use one of your application that you already installed in your mobile phone and when this application starts sensing or collecting your information and your phone is turned off. So in this case, the sensing process is stopped because it is done without the knowledge of the users and it is based on the phone conditions.
Also for participatory, it's more expensive or require a high cost than the opportunistic one for the same reason, because in a participatory, as we say that users, they volunteer to participate. While this is not the case for opportunistic one and those users are sometimes paid for this kind of works, you know, for collecting the information. For this reason, it's it is more expensive, expensive than opportunistic crowd sensing.
So here, this is some example of mobile crowd sensing applications. Like, for example, the way now YouTube, Facebook, Uber, Upwork, Instacart and so on that deploy a set of workers and those workers are paid based on their job. So here, this difference in companies, IoT devices, technologies, this is what we call heterogeneous. So a heterogeneous environment is an environment that deploy different or diverse, diverse devices or transmission technologies.
For example, there is some devices that can share, for example, we can share our pictures using, for example, Wi-Fi directs or Bluetooth or even using Wi-Fi or 4G. And this is a different, completely different transmission technology. And also we can share our information using smartphone as well as using computer or, for example, dash camera, camera surveillance and so on.
So this diversity, this is what we call heterogeneous environment. So in this kind of environment, also the data is heterogeneous. Why? Because it depends on the device conditions and the environment conditions and also the transmission technology conditions. So the generated data, after being collected or sensitive, it will be gathered and send it to a centralized platform. At this centralized platform, this data is stored and requires some additional computation and additional processing and filtering.
So this centralized platform requires some or has some challenges or limitations, for example, delay and answering real time requirements. Because, as we say, we have a very big or a very big amount of data that is generated, a different data, heterogeneous data that is also stored. And this kind of information need or require more time to be processed and to be ready to send to the end users.
So it takes too much time to give a feedback to answer a specific request from a specific user. Another challenge is discovering the non-connected devices. So what we mean by non-connected devices, it's, for example, for smartphone or for other devices, we can use Wi-Fi or 4G communication network. So, in this case, those devices are accessible by the server and discovered by the server.
While, for example, in case of sharing using Wi-Fi direct or Bluetooth communication, those devices can be hidden from the server. So this is one of the main limitations of a centralized platform. Also, they are complex on computing the data due to the volume of the data that is stored there and that is generated. And the last one is the high cost.
They are very expensive. So the centralized platform are very expensive because for the same reasons, they require a strong memory to store the whole information and also require a very high capacity of processing. So, to overcome this, some of the limitations of the centralized platform, there is two solutions that was recently proposed. The first one is using a decentralized or a distributed architecture.
And this distributed architecture, as you can distinguish in these two figures, the first one, the whole information is sent to the cloud. So, all the information, whatever, the source, are sent to a centralized platform. While in the distributed architecture, the information is distributed according to the area of interest, according to the location of the server, and according to the location of the workers or the users itself.
So, here, the information, instead of going to a centralized platform, it is distributed in different places or in different services. And this makes the processing of this data very easy and more fast than the case for a cloud or a centralized platform. So, the second solution is the edge computing. So, the edge computing is considered an enabler for IoT services that allow computing, that is an optimal solution for computing a large amount of data.
So, it is used like an intermediate layer between the cloud server and the mobile users. So, it is placed in between. So, it allows to reduce the complexity of computation because it is placed near to the end users. And also, it is responsible to do some data filtering, aggregation, processing, and storage, and respond in real time to the final users.
So, this edge computing has a set of benefits. The first one is the parallel and the partitioning problem space based on the location. As we said, Yannir, in the previous figures, that there is a set of edge servers, and each one of them collect a set of information that is situated in the same location. And those information are processed in parallel.
So, all the servers work at the same time on the same, not same information, but on different information, but they are complementary information. And after that, after this computation, if, for example, the information or the request that is sent by the user, it's a very easy task or it can be answered in real time, the edge node is responsible to answer to that user.
But in case if the user needs more information, this, the edge computer sends the request to the cloud computing, and the response will be coming back from the cloud computing. But, and for small tasks, the answer or the interaction between the edge server and the mobile user, it's direct. So, another benefit, so this interaction or being placed near to the end user provide other benefits, like reducing or flooding the computation for both mobile devices and cloud servers, reduce the computational complexity of a centralized cloud, and decrease the latency latency and the privacy treat.
So, for offloading computation, the edge computing, as we say, is placed in intermediate between the mobile devices and cloud server. So, for example, for example, in a mobile devices, if the data is collected, for example, if I'm using my smartphone and the data is collected, as we can know that a smartphone has limited memory, storage memory and capacity of processing.
So, for example, if I collect a very huge number of information, I cannot process it or compute it in my own smartphone. So, what I do, I send it to the edge computer. This can edge computer do this computation. So, in this case, I offload the task that normally should be done in a smartphone. It is already done in edge computer.
And also, for example, if this task, usually it is processed on a cloud computer, but in case of deploying edge computer, this task can be treated locally in the edge computer without even going to the cloud and getting a feedback from the cloud. So, it's like a double benefit of loading the smartphone and also loading the cloud servers.
And for the latency, this is for sure near to the end users. So, it means that there is a fast answer or fast replay to a specific request. It means that there is a low latency. For the privacy trees, for example, if all the information, and this is the case of centralized platform, if all the information is stored on a cloud or a centralized platform, for example, in case of attacks or, for example, material destroyed or crash or something like this, all the information will be lost.
So, in case of distributed architecture, the information is distributed in different places. And for example, if some of this information is missing, so, based on what's still as information in the other servers, we can, for example, predict what is missing or we can estimate or expect, you know, the missing information. So, and this is what reduced or increased, you know, reduced the privacy tree.
So, here, based on what we learned about cloud sensing, about the limitation of centralized platform and about the benefit of each computer and our centers, we try to deploy or to make like a combination of all these concepts and one architecture. So, we proposed an off-site solution, which is an edge computer-based solution that deploy edge nodes, different entities of edge nodes, and it's a distributed architecture that deploy a set of servers, a set of edge nodes, and select a set of workers, which is the third concept.
So, it's deployed in a crowd sensing environment for worker selection or worker recruitment. So, here, I will show you. Yeah. So, here, our solution addresses the following concepts. First, the resource discovery, because what we assume that our architectures not only deploy the connected devices, the devices that use Wi-Fi or 4G or 5G, in general, wireless communications, but also, it can discover other non-connected devices, for example, that deploy the Bluetooth.
So, because our solution aims to integrate the concept of heterogeneous environments. So, here, we use a different technology. And also, it's a task allocation, because it's based on the task received from the server or the worker also, because we assume also that the task can be initiated by the different entities. Maybe a response to a request from an edge server, or for example, for an event detection, it will be initiated by the workers.
So, it's a context where, because it's based on what we have in the environment, what we have in terms of IoT devices, the connected and the non-connected devices, also the number of participants or number of workers available on an area of interest, their location, the location of task, and so on. So, all these kinds of requirements should be considered, are considered in our solution.
Also, heterogeneous worker selection, so heterogeneous IoT devices and heterogeneous wireless or transmission technology lead us to address the concept of heterogeneous workers. And finally, the budget constraints. Why? We introduced the concept of budget here, because this is one of the main parameters of crowdsensing. So, crowdsensing deploy first sets of workers, mobile devices, as we say, smartphone or wheelable devices.
And in case of participatory crowdsensing, we need a specific budget because the workers that are recruited in this kind of crowdsensing, they are already paid. So, it's according to the budget that we assume or that we attribute to each node, the selection will be stopped. If, for example, if that budget is rich, so we stop the recruitment.
But if it is not the case, we keep recruitment, for sure. So, here, this is just a video that explains clearly how the solution works. See? Okay. So, as we can see here, the final task is the worker selection. It's attributed to each LAN or local edge node in each cluster. So, based on two parameters, which are the quality of services provided by each worker.
So, each LAN or each local edge node selects the best workers based on their quality of services and also based on budget. Because this is why we say that our solution is a budget constraint. So, according to a viable budget, we keep recruiting the workers. So, as a result or a finding, we found that our solution selects best workers.
And based on that provide a very high quality of services and the quality of services include devices characteristics and also workers' reputation. So, the workers selected by our solution have a high reputation compared to the centralized one. So, as a takeaway, all what we need to know is that mobile crowd sensing is a human-driven paradigm that has as a main focus.
It's collecting data from a physical environment using mobile devices. And a centralized mobile cloud sensing, it suffers from a set of limitations that can be overcome by the deployment of mobile edge computing. And without forgetting the concept of heterogeneous environment, which is deploying a diverse entities or devices that have a different character and content. And if you want to connect and contact me, this is my LinkedIn, my Gmail and Google Scholar and Scope if you want to read some of my resources.
And thank you so much.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
Why mobile crowdsensing needs edge computing, not one central cloud
Hanane Lamaazi shows how a distributed edge architecture answers faster, finds hidden devices and recruits better workers within a budget.
In short
Mobile crowdsensing gathers data through the smartphones and wearables people carry, but sending all of it to a single centralized platform creates delays, misses devices, overloads computation and costs a lot. Hanane Lamaazi, a Post-Doctoral Fellow at Khalifa University, argues that a distributed architecture built on mobile edge computing solves these problems, and presents her team's solution for selecting crowdsensing workers by quality of service under a budget.
Key takeaways
Edge computing technology
Distributed architecture
Worker selection in mobile crowdsensing
From the Internet of Things to crowdsensing
Lamaazi opens with two questions for the audience: whether they know their personal information is useful to third parties, and whether they know where and how it is stored. She then places crowdsensing inside the wider Internet of Things, which connects everyday objects across healthcare, home automation, transportation and industry. As the speaker cited from IoT Analytics, 1,414 IoT projects were tracked in 2020, with manufacturing and industrial projects the most common at 22%, followed by transportation and energy.
Crowdsensing, she explains, collects data about a phenomenon by deploying "workers" who carry mobile devices and perform specific tasks. She separates two kinds. In participatory crowdsensing, users volunteer and decide what, when and where to sample; they are often paid, which makes it more expensive. Opportunistic crowdsensing is different:
"The data is sensed, collected, and shared automatically and sometimes even without the knowledge of the user." (3:52)
Because opportunistic sensing depends on the phone's condition, it stops if the app crashes or the phone switches off. Lamaazi points to services such as YouTube, Facebook, Uber, Upwork and Instacart as examples of platforms that deploy paid workers, and introduces the idea of a heterogeneous environment: many kinds of devices, transmission technologies (Wi-Fi, Wi-Fi Direct, Bluetooth, 4G) and therefore many kinds of data.
What goes wrong with a centralized platform
In the usual setup, all that heterogeneous data is gathered and sent to one centralized platform for storage, filtering and processing. Lamaazi lists four limitations. The first is delay:
"So it takes too much time to give a feedback to answer a specific request from a specific user." (8:29)
The second is discovery. Devices on Wi-Fi or 4G are visible to the server, but others are not:
"While, for example, in case of sharing using Wi-Fi direct or Bluetooth communication, those devices can be hidden from the server." (9:03)
The third is the complexity of computing over such a large volume of data, and the fourth is cost, since a central platform needs strong memory and very high processing capacity.
Distributing the work and moving it to the edge
Two recently proposed answers address these limits. A distributed architecture spreads information according to the area of interest, the location of servers and the location of workers, which makes processing easier and faster. Edge computing adds a layer close to users:
"So, it is used like an intermediate layer between the cloud server and the mobile users." (11:17)
Edge servers filter, aggregate, process and store data, and respond to users in real time. Lamaazi describes the benefits. Edge servers in different locations work in parallel on complementary data. Small requests are answered directly by the edge node, while larger ones go to the cloud. Computation is offloaded from both ends: a phone with limited memory and processing can send its data to an edge server, and tasks that would normally go to the cloud can be handled locally.
"So, it's like a double benefit of loading the smartphone and also loading the cloud servers." (14:43)
Proximity also lowers latency. And because data is spread across servers, an attack or crash does not destroy everything at once; missing information can be estimated from what other servers still hold.
An edge-based solution for selecting workers
Lamaazi's team combined these ideas into a single edge-based, distributed architecture for worker selection in crowdsensing. It addresses resource discovery, including non-connected devices such as those using Bluetooth; task allocation, where tasks can come from a server or be initiated by workers, for example during event detection; context awareness about devices, workers and locations; heterogeneous worker selection; and budget constraints, since paid participatory workers mean recruitment must stop once the budget is reached.
"So, each LAN or each local edge node selects the best workers based on their quality of services and also based on budget." (21:16)
Quality of service covers both device characteristics and worker reputation. The finding she reports from this research:
"So, the workers selected by our solution have a high reputation compared to the centralized one." (21:45)
She closes by restating that mobile crowdsensing is a human-driven paradigm for collecting data from the physical environment with mobile devices, and that the limits of centralized platforms can be overcome by mobile edge computing, without forgetting the reality of heterogeneous environments.
Watch the talk
Mobile edge computing-based distributed architecture in mobile crowdsensing, Hanane Lamaazi, 23 minutes: https://www.womentech.net/video/mobile-edge-computing-based-distributed-architecture-in-mobile-crowdsensing
About the speaker
Hanane Lamaazi is a Post-Doctoral Fellow at Khalifa University, holding a PhD in computer sciences and networks, with research interests in the Internet of Things, the RPL routing protocol, edge computing and crowdsensing. She is a TPC member for several scientific conferences and a reviewer for several scientific journals.
Questions this talk answers
What is the difference between participatory and opportunistic crowdsensing?
Why do centralized mobile crowdsensing platforms struggle?
What role does edge computing play between phones and the cloud?
How does a distributed architecture reduce the risk of losing data?
How does the proposed solution choose crowdsensing workers?
Drafted with AI from the talk transcript. Editor review pending.