Edge Computing
Edge Computing
Blesson Varghese, Nan Wang, Sakil Barbhuiya, Peter Kilpatrick and Dimitrios S. Nikolopoulos
School of Electronics, Electrical Engineering and Computer Science
Queen’s University Belfast, UK
Email: {varghese, nwang03, sbarbhuiya03, p.kilpatrick, d.nikolopoulos}@qub.ac.uk
I. I NTRODUCTION
The article ‘Above the clouds’1 presented challenges and
opportunities in cloud computing [1]. Cloud research since
has rapidly progressed paving the way for many competitors
in a crowded marketplace. Although the vision of offering
computing as a utility was achieved [2], the research space is
Fig. 1. Edge devices and edge nodes in relation to the cloud
still far from saturation and offers interesting opportunities.
Many cloud applications are user-driven, which has resulted
in opportunities for large-scale data analytics. However, using
the cloud as a centralised server simply increases the frequency
of communication between user devices, such as smartphones,
tablets, wearables and gadgets, we refer to as edge devices,
and geographically distant cloud data centres. This is limiting
for applications that require real-time response. Hence, there
has been a need for looking ‘beyond the clouds’ towards the
edge of the network as shown in Figure 1, we refer to as
edge computing [3], [4], but is also known as fog computing
[5], [6] or cloudlet computing [7], [8]. The aim is to explore
possibilities of performing computations on nodes through
which network traffic is directed, such as routers, switches and
base stations, we refer to as edge nodes. The objective of this
paper is to define the motivation, challenges and opportunities
in edge computing, which is summarised in Figure 2. Fig. 2. Motivation, challenges and opportunities in edge computing
II. M OTIVATION
We have identified the following five needs that motivate be generalised for any web-based application [9]. Location-
computing on edge nodes. aware applications such as Foursqure2 and Google Now3 are
becoming popular among mobile users. Computing on edge
1) Decentralised Cloud and Low Latency Computing nodes closer to application users could be exploited as a
Centralised cloud computing may not always be the best platform for application providers to improve their service.
strategy for applications that are geographically distributed. Numerous data streams are generated by edge devices and
Computing needs to be performed closer to the source of the making real-time decisions is not possible when analytics is
data to improve the service that is delivered. This benefit can
2 https://foursquare.com
1 https://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-28.pdf 3 https://www.google.co.uk/landing/now/
performed on a distant cloud. Real-time applications, such 4) Dealing with Data Explosion and Network Traffic
as a visual guiding service using a wearable camera have The number of edge devices is growing at an enormous rate
a preferred response time between 25ms to 50ms [10]. The and one-third of the world’s population is expected to have a
use of current cloud infrastructures poses a serious latency smart phone by 201878 . Consequently, the volume of data that
challenge between an edge device and the cloud. The round will be generated will also increase9 ; it is anticipated that 43
trip time from Canberra to Berkeley is approximately 175ms trillion gigabytes of data will be generated in 202010 . This
[7], which is far from requirements of latency-sensitive appli- places the need for expanding data centers to support moni-
cations. Video streaming is the largest mobile traffic generator4 toring and analytical workloads, but this again raises concerns
and satisfying users remains challenging if the bulk of this with regard to sustainable energy consumption of data centers.
traffic needs to be channelled to the same source. Similarly, There are attempts to mitigate the energy challenge by per-
multimedia applications, such as on-demand gaming on cur- forming analytics on the edge device [16], [17], [18]. However,
rent cloud infrastructure, pose similar latency issues for the this is restrictive due to resource limitations in the edge devices
gamer [11]. Here, to complement the computations performed and collective analytics (of multiple edge devices) cannot be
on the cloud, edge nodes that are located closer (for example, realistically done on an edge device. Another concern with
routers or base stations one-hop away from an edge device) increasing data generation is the volume of network traffic
to users can be leveraged to reduce network latency. to a central server or cloud thereby reducing the response
2) Surmounting Resource Limitations of Front-end Devices time of edge devices. This presents the potential of using
nodes that are a hop away in the network for complementing
User devices, such as smart phones, have relatively re-
computations of the device or of the data center to cope with
stricted hardware resources when compared to a server in
the growth of data as well as distributing traffic in a network.
a data centre. These front-end devices capture sensory input
in the form of text, audio, video, touch or motion and are 5) Smart Computation Techniques
processed by a service provided by the cloud [12]. The Data generated at a user end needs to be often transported to
front-end devices cannot perform complex analytics due to a cloud server for performing any meaningful analytics which
middleware and hardware limitations [13]; this may sometimes has obvious latency and energy implications. However, there
be possible at the expense of draining the battery. Hence, often is potential to harness resources at the far end of the network
data needs to be sent to the cloud to meet the computational if computation can be hierarchically distributed [19]. For
demands of processing data and meaningful information is example, a typical application pipeline may initially filter data
then relayed back to the front-end. However, not all data generated on the device, after which analytical workloads are
from a front-end device will need to be used by the service executed on the edge nodes through which data is transmitted,
to construct analytical workloads on the cloud. Potentially, before finally arriving at the cloud server where more complex
data can be filtered or even analysed at edge nodes, which tasks are performed. Alternatively, it may be possible for data
may have spare computational resources to accommodate data centres to offload computations requiring limited resources
management tasks. on to edge nodes, or for the edge nodes to make use of
3) Sustainable Energy Consumption volunteer devices to enhance computational capabilities [20].
Edge nodes can facilitate computations nearer to the source of
There is a significant body of research that has investigated
data (or where data is generated) and can incorporate strategies
the energy consumption of cloud data centres [14], [15]. Data
for remotely enhancing capabilities of front-end devices.
centres in the next decade are likely to consume three times
as much energy consumed today 5 and there is a greater need III. C HALLENGES
for adopting energy efficient strategies that can minimise the
Edge computing is still in its infancy and a framework to
energy usage6 . With more and more applications moving on
facilitate this is not yet available. Such frameworks will need
to the cloud it may become untenable to meet the increasing
to satisfy requirements, such as application development to
energy demands. The challenge of increasing energy could
process requests in real-time on edge nodes. Current cloud
be alleviated in a small proportion by incorporating sensible
computing frameworks, such as the Amazon Web Service11 ,
power management strategies in that a number of analytical
Microsoft Azure12 and Google App Engine13 , can support
tasks are performed on edge nodes, such as base stations or
data-intensive applications, but implementing real-time data
routers, closer to the data source instead of overloading data
centres with trivial tasks that could perhaps be performed on 7 http://www.pewglobal.org/2016/02/22/smartphone-ownership-and-
are-the-new-polluters.html 13 https://appengine.google.com/
processing at the edge of the network is still an open research through a variety of techniques that are incorporated into mon-
area [21], [22]. Additionally, the requirement of deploying itoring tools [24], [25], [26], [27] and service brokerages [28],
application workloads on edge nodes will need to be well [29], [30], [31]. Techniques such as benchmarking underpin
understood. Deployment strategies - where to place a work- decision-making for mapping tasks onto the most suitable
load, connection policies - when to use edge nodes and resources for improving performance.
heterogeneity - how to deal with different types of nodes need However, exploiting the edge of the network requires dis-
to be taken into account for deploying applications on the covery mechanisms to find appropriate nodes that can be lever-
edge. For achieving such a framework, we envisage that the aged in a decentralised cloud set up. These mechanisms cannot
following five research challenges at the hardware, middleware be simply manual due to the sheer volume of devices that will
and software layer will need to be addressed. be available at this layer. Moreover, they will need to cater for
heterogeneous devices from multiple generations as well as
Challenge 1 - General Purpose Computing on Edge Nodes modern workloads, for example large scale machine learning
In theory, edge computing can be facilitated on several tasks, which were previously not considered. Benchmarking
nodes that are located between the edge device and the methods will need to be significantly rapid in making known
cloud, including access points, base stations, gateways, traffic the availability and capability of resources. These mechanisms
aggregation points, routers, switches, etc. Base stations, for must allow for seamless integration (and removal) of nodes
example, incorporate Digital Signal Processors (DSPs) that in the computational workflow at different hierarchical levels
are customised to the workloads they handle. In practice, without increasing latencies or compromising the user experi-
base stations may not be suitable for handling analytical ence. Reliably and proactiviely dealing with faults on the node
workloads simply because DSPs are not designed for general and autonomically recovering from them will be desirable.
purpose computing. Moreover, it is not readily known if these Existing methods used in the cloud will not be practical in
nodes can perform computations in addition to their existing this context for the discovery of edge nodes.
workloads. The OCTEON Fusion®Family14 by CAVIUM,
Challenge 3 - Partitioning and Offloading Tasks
a small cell ”Base Station-on-a-chip” family, scales from
6 to 14 cores to support users ranging from 32 to 300+. Evolving distributed computing environments have resulted
Such base stations could perhaps be used during off-peak in the development of numerous techniques to facilitate par-
hours to exploit the computational capabilities of multiple titioning of tasks that can be executed at multiple geographic
computing cores available. A number of commercial vendors locations [32], [33]. For example, workflows are partitioned for
have taken a first step to realise edge computing using software execution in different locations [34], [35]. Task partitioning is
solutions. For example, Nokia’s software solution15 for mobile usually expressed explicitly in a language or management tool.
edge computing (MEC) aims to enable base station sites for However, making use of edge nodes for offloading compu-
edge computing. Similarly, Cisco’s IOx16 offers an execution tations poses the challenge of not simply partitioning com-
environment for its integrated service routers. These solutions putational tasks efficiently, but doing this in an automated
are specific to hardware and hence may not be suitable in manner without necessarily requiring to explicitly define the
a heterogeneous environment. One challenge in the software capabilities or location of edge nodes. The user of a language
space will be to develop solutions that are portable across that can leverage edge nodes may anticipate flexibility to
different environments. define a computation pipeline - hierarchically in sequence (first
There is research in upgrading the resources of edge nodes at the data centre then at the edge nodes or first at the edge
to support general purpose computing. For example, a wireless node and then at the data centre) or potentially over multiple
home router can be upgraded to support additional workloads edge nodes simultaneously. Inherently, there arises the need
[23]. Intel’s Smart Cell Platform17 uses virtualisation for for developing schedulers that deploy partitioned tasks onto
supporting additional workloads. Replacing specialised DSPs edge nodes.
with comparable general purpose CPUs gives an alternative
Challenge 4 - Uncompromising Quality-of-Service (QoS) and
solution but this requires a huge investment.
Experience (QoE)
Challenge 2 - Discovering Edge Nodes Quality delivered by the edge nodes can be captured by QoS
and quality delivered to the user by QoE. One principle that
Discovering resources and services in a distributed com-
will need to be adopted in edge computing is to not overload
puting environment is an area that is well explored. This is
nodes with computationally intensive workloads [36], [37].
facilitated in both tightly and loosely coupled environments
The challenge here is to ensure that the nodes achieve high
14 http://www.cavium.com/OCTEON-Fusion.html throughput and are reliable when delivering for their intended
15 http://networks.nokia.com/portfolio/solutions/mobile-edge- workloads if they accomodate additional workloads from a
computing#tab-highlights data center or from edge devices. Regardless of whether an
16 http://www.cisco.com/c/en/us/products/cloud-systems-
edge node is exploited, the user of an edge device or a data
management/iox/index.html
17 http://www.intel.com/content/www/us/en/communications/smart-cells- centre expects a minimum level of service. For example, when
revolutionize-service-delivery.html a base station is overloaded, it may affect the service provided
to the edge devices that are connected to the base station. A (ISO)23 , Cloud Standards Customer Council (CSCC)24 and
thorough knowledge of the peak hours of usage of edge nodes the International Telecommunication Union (ITU)25 . However,
is required so that tasks can be partitioned and scheduled such standards will now need to be reconsidered in light of ad-
in a flexible manner. The role of a management framework ditional stakeholders, such as public and private organisations
will be desirable but raises issues related to monitoring, that own edge nodes, to define the social, legal and ethical
scheduling and re-scheduling at the infrastructure, platform aspects of using edge nodes. This is certainly not an easy
and application levels18 . task and requires commitment and investment from public and
private organisations and academic institutions.
Challenge 5 - Using Edge Nodes Publicly and Securely Standards can be implemented only if the performance
Hardware resources that are owned by data centres, su- of edge nodes can be reliably benchmarked against well
percomputing centres and private organisations using virtu- known metrics. Benchmarking initiatives for the cloud include
alisation can be transformed to offer computing as a utility. those by the Standard Performance Evaluation Corporation
The associated risks for a provider and users are articulated19 , (SPEC)26 and by numerous academic researchers [40], [41],
thereby offering computing on a pay-as-you-go basis. This has [42], [43]. In a noisy environment, such as the cloud, bench-
resulted in a competitive marketplace with numerous options marking poses significant challenges. The current state-of-the-
and choices to satisfy computing consumers by meeting Ser- art is not yet mature and significant research is required to
vice Level Agreements (SLAs) [38]. deliver comprehensive benchmarking suites that can gather
However, if alternative devices, such as switches, routers metrics accurately. Benchmarking edge nodes will therefore
and base stations, need to be used as publicly accessible edge be more challenging, but opens new avenues of research.
nodes a number of challenges will need to be addressed. Using edge nodes is an attractive prospective when respon-
Firstly, the risk associated by public and private organisations sibilities, relationships and risks are defined. Similar to a cloud
that own these devices as well as those that will employ marketplace, an edge computing marketplace that offers a
these devices will need to be articulated. Secondly, the in- variety of edge nodes on a pay-as-you-go basis is feasible.
tended purpose of the device, for example, a router managing Research in defining SLAs for edge nodes and pricing models
internet traffic, cannot be compromised when used as an will be required to create such a marketplace.
edge computing node. Thirdly, multi-tenancy on edge nodes
Opportunity 2 - Frameworks and Languages
will only be possible with technology that places security
as a prime concern. Containers, for example, a potential There are many options to execute applications in the cloud
lightweight technology usable on edge nodes will need to paradigm. In addition to popular programming languages,
demonstrate more robust security features [39]. Fourthly, a there is a wide variety of services to deploy applications on
minimum level of service will need to be guaranteed to a user the cloud. When resources outside the cloud are employed,
of the edge node. Fifthly, the workloads, computation, data for example, running a bioinformatics workload on the public
location and transfer, cost of maintenance and energy bills will cloud where input data is obtained from a private database,
need to be considered for developing suitable pricing models a workflow is usually employed. Software frameworks and
to make edge nodes accessible. toolkits for programming large workflows in a distributed
environment is a well defined research avenue [44], [45].
IV. O PPORTUNITIES However, with the addition of edge nodes that may support
general purpose computing, there will be the need for develop-
Despite challenges that arise when realising edge comput-
ing frameworks and toolkits. The use cases for edge analytics
ing, there are numerous opportunities for academic research.
are likely to differ from existing workflows, which are mostly
We identify five such opportunities.
explored for scientific domains, such as bioinformatics [46]
Opportunity 1 - Standards, Benchmarks and Marketplace or astronomy [47]. Given that edge analytics will find its use-
cases in user-driven applications, the existing frameworks may
Edge computing can be realised in practice and be made not be appropriate to express an edge analytics workflow. The
accessible publicly if responsibilities, relationships and risks programming model that aims to exploit edge nodes will need
of all parties involved are articulated. There are numerous to support task and data level parallelism and at the same time
efforts to define a variety of cloud standards, such as by Na- execute workloads on multiple hierarchical levels of hardware.
tional Institutes of Standards and Technology(NIST)2021 , IEEE The language that supports the programming model will need
Standards Association22 , International Standards Organisation to take into account the heterogeneity of hardware and the
18 https://portal.etsi.org/portals/0/tbpages/mec/docs/mobile- capacity of resources in the workflow. If edge nodes are more
edge computing - introductory technical white paper v1%2018-09-14.pdf vendor specific, then the frameworks supporting the workflow
19 http://www.cloud-council.org/deliverables/CSCC-Practical-Guide-to-
Cloud-Service-Agreements.pdf 23 http://www.iso.org/iso/home/store/catalogue tc/catalogue tc browse.htm?
20 http://www.nist.gov/itl/cloud/ commid=601355
21 http://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800- 24 http://www.cloud-council.org/
145.pdf 25 http://www.itu.int/en/ITU-T/jca/Cloud/Pages/default.aspx
22 http://cloudcomputing.ieee.org/standards 26 https://www.spec.org/benchmarks.html#cloud
will need to account for it. This is more complex than existing and researchers cannot gain access to infrastructure owned
models that make the cloud accessible. by the industry or government to validate and refine their
research [52]. Premier academic institutions that have reliable
Opportunity 3 - Lightweight Libraries and Algorithms industry and government relations have however produced
Unlike large servers edge nodes will not support heavy- more meaningful and impactful research [53].
weight software due to hardware constraints. For instance, a Research in the edge computing space can be driven by an
small cell base station with Intel’s T3K Concurrent Dual-Mode open consortium of industry partners, such as mobile operators
system-on-chip (SoC)27 has a 4-core ARM-based CPU and and developers, gadget developers and cloud providers, as well
limited memory, which will not be sufficient for executing as interested academic partners to the mutual benefit of both.
complex data processing tools such as Apache Spark28 that
V. C ONCLUSION
requires at least 8 cores CPU and 8 gigabyte memory for good
performance. Edge analytics require lightweight algorithms Edge computing is still in its infancy and has potential to
that can do reasonable machine learning or data process- pave the way for more efficient distributed computing. This
ing tasks [48], [49]. Apache Quarks29 , for example, is a paper has highlighted the significant potential of computing at
lightweight library that can be employed on small footprint the edge of the network and presented five research challenges
edge devices such as smart phones to enable real-time data and five rewarding opportunities in realising edge computing.
analytics. However, Quarks supports basic data processing, R EFERENCES
such as filtering and windowed aggregates, which are not
[1] M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz, A. Konwinski,
sufficient for advanced analytical tasks such as context-aware G. Lee, D. Patterson, A. Rabkin, I. Stoica, and M. Zaharia, “A View
recommendations. Machine learning libraries that consume of Cloud Computing,” Communications of the ACM, vol. 53, no. 4, pp.
less memory and disk usage would benefit data analytical tools 50–58, 2010.
[2] R. Buyya, C. S. Yeo, S. Venugopal, J. Broberg, and I. Brandic, “Cloud
for edge nodes. TensorFlow30 is another example framework Computing and Emerging IT Platforms: Vision, Hype, and Reality for
that supports deep learning algorithms and supports heteroge- Delivering Computing as the 5th Utility,” Future Generation Computer
neous distributed systems, but its potential for edge analytics Systems, vol. 25, no. 6, pp. 599 – 616, 2009.
[3] P. Garcia Lopez, A. Montresor, D. Epema, A. Datta, T. Higashino,
is still to be explored. A. Iamnitchi, M. Barcellos, P. Felber, and E. Riviere, “Edge-centric
Computing: Vision and Challenges,” SIGCOMM Computer Communi-
Opportunity 4 - Micro Operating Systems and Virtualisation cation Review, vol. 45, no. 5, pp. 37–42, Sep. 2015.
[4] M. Satyanarayanan, P. Simoens, Y. Xiao, P. Pillai, Z. Chen, K. Ha,
Research in micro operating systems or microkernels can W. Hu, and B. Amos, “Edge Analytics in the Internet of Things,” IEEE
provide inroads to tackling challenges related to deployment Pervasive Computing, vol. 14, no. 2, pp. 24–31, Apr 2015.
[5] T. H. Luan, L. Gao, Z. Li, Y. Xiang, and L. Sun, “Fog Computing:
of applications on heterogeneous edge nodes. Given that these Focusing on Mobile Users at the Edge,” CoRR, vol. abs/1502.01815,
nodes do not have substantial resources like in a server, the 2015. [Online]. Available: http://arxiv.org/abs/1502.01815
general purpose computing environment that is facilitated on [6] F. Bonomi, R. Milito, J. Zhu, and S. Addepalli, “Fog Computing and
Its Role in the Internet of Things,” in Proceedings of the Workshop on
the edge will need to exhaust fewer resources. The benefits Mobile Cloud Computing, 2012, pp. 13–16.
of quick deployment, reduced boot up times and resource [7] M. Satyanarayanan, P. Bahl, R. Caceres, and N. Davies, “The Case for
isolation are desirable [50]. There is preliminary research VM-Based Cloudlets in Mobile Computing,” IEEE Pervasive Comput-
ing, vol. 8, no. 4, pp. 14–23, 2009.
suggesting that mobile containers that multiplex device hard- [8] K. Gai, M. Qiu, H. Zhao, L. Tao, and Z. Zong, “Dynamic Energy-aware
ware across multiple virtual devices can provide similar per- Cloudlet-based Mobile Cloud Computing Model for Green Computing,”
formance to native hardware [51]. Container technologies, Journal of Network and Computer Applications, vol. 59, no. C, pp. 46–
54, 2016.
such as Docker31 are maturing and enable quick deployment [9] J. Zhu, D. S. Chan, M. S. Prabhu, P. Natarajan, H. Hu, and F. Bonomi,
of applications on heterogeneous platform. More research is “Improving Web Sites Performance Using Edge Servers in Fog Com-
required to adopt containers as a suitable mechanism for puting Architecture,” in Proceedings of the International Symposium on
Service Oriented System Engineering, 2013, pp. 320–323.
deploying applications on edge nodes. [10] S. Agarwal, M. Philipose, and P. Bahl, “Vision: The Case for Cellular
Small Cells for Cloudlets,” in Proceedings of the International Workshop
Opportunity 5 - Industry–Academic Collaborations on Mobile Cloud Computing & Services, 2014, pp. 1–5.
[11] S. Choy, B. Wong, G. Simon, and C. Rosenberg, “The Brewing Storm in
Edge computing offers a unique opportunity for academia Cloud Gaming: A Measurement Study on Cloud to End-user Latency,”
to re-focus its research activities broadly in applied distributed in Proceedings of the Workshop on Network and Systems Support for
computing, specifically within cloud and mobile computing. It Games, 2012.
[12] D. Damopoulos, G. Kambourakis, and G. Portokalidis, “The Best of
is not easy for academic research to focus on scale without Both Worlds: A Framework for the Synergistic Operation of Host and
making assumptions that may not always correspond to reality. Cloud Anomaly-based IDS for Smartphones,” in Proceedings of the
This is simply because a large number of academic institutions European Workshop on System Security, 2014, pp. 6:1–6:6.
[13] A. Houmansadr, S. A. Zonouz, and R. Berthier, “A Cloud-based Intru-
27 http://www.intel.com/content/dam/www/public/us/en/documents/solution- sion Detection and Response System for Mobile Phones,” in Proceedings
of the IEEE/IFIP International Conference on Dependable Systems and
briefs/transcede-t3k-solution-brief.pdf Networks Workshops, 2011, pp. 31–32.
28 http://spark.apache.org
[14] A. Beloglazov, J. Abawajy, and R. Buyya, “Energy-aware Resource
29 http://quarks.incubator.apache.org
Allocation Heuristics for Efficient Management of Data Centers for
30 https://www.tensorflow.org/
Cloud Computing,” Future Generation Computer Systems, vol. 28, no. 5,
31 https://www.docker.com pp. 755–768, May 2012.
[15] A. Greenberg, J. Hamilton, D. A. Maltz, and P. Patel, “The Cost of [34] W. Chen and E. Deelman, “Integration of Workflow Partitioning and
a Cloud: Research Problems in Data Center Networks,” SIGCOMM Resource Provisioning,” in Proceedings of the IEEE/ACM International
Computer Communication Review, vol. 39, no. 1, pp. 68–73, Dec. 2008. Symposium on Cluster, Cloud and Grid Computing, 2012, pp. 764–768.
[16] M. Sun, M. Zheng, J. C. S. Lui, and X. Jiang, “Design and imple- [35] W. Tang, J. Jenkins, F. Meyer, R. Ross, R. Kettimuthu, L. Winkler,
mentation of an android host-based intrusion prevention system,” in X. Yang, T. Lehman, and N. Desai, “Data-Aware Resource Scheduling
Proceedings of the Annual Computer Security Applications Conference, for Multicloud Workflows: A Fine-Grained Simulation Approach,” in
2014, pp. 226–235. Proceedings of the IEEE International Conference on Cloud Computing
[17] A. Shabtai, U. Kanonov, Y. Elovici, C. Glezer, and Y. Weiss, “An- Technology and Science, 2014, pp. 887–892.
dromaly: A Behavioral Malware Detection Framework for Android [36] M. T. Beck and M. Maier, “Mobile Edge Computing: Challenges for
Devices,” Journal of Intelligent Information Systems, vol. 38, no. 1, pp. Future Virtual Network Embedding Algorithms,” in Proceedings of
161–190, Feb. 2012. the International Conference on Advanced Engineering Computing and
[18] G. Dini, F. Martinelli, A. Saracino, and D. Sgandurra, “MADAM: A Applications in Sciences, 2014, pp. 65–70.
Multi-level Anomaly Detector for Android Malware,” in Proceedings [37] P. Simoens, L. Van Herzeele, F. Vandeputte, and L. Vermoesen, “Chal-
of the International Conference on Mathematical Methods, Models lenges for Orchestration and Instance Selection of Composite Services in
and Architectures for Computer Network Security: Computer Network Distributed Edge Clouds,” in Proceedings of the IFIP/IEEE International
Security, 2012, pp. 240–253. Symposium on Integrated Network Management, 2015, pp. 1196–1201.
[38] S. A. Baset, “Cloud SLAs: Present and Future,” ACM SIGOPS Operating
[19] P. Hari, K. Ko, E. Koukoumidis, U. Kremer, M. Martonosi, D. Ottoni,
Systems Review, vol. 46, no. 2, pp. 57–66, 2012.
L.-S. Peh, and P. Zhang, “SARANA: Language, Compiler and Run-
[39] T. Bui, “Analysis of Docker Security,” CoRR, vol. abs/1501.02967,
time System Support for Spatially Aware and Resource-aware Mobile
2015. [Online]. Available: http://arxiv.org/abs/1501.02967
Computing,” Philosophical Transactions of the Royal Society of London
[40] B. Varghese, L. Subba, L. Thai, and A. Barker, “DocLite: A Docker-
A: Mathematical, Physical and Engineering Sciences, vol. 366, no. 1881,
Based Lightweight Cloud Benchmarking Tool,” in Proceedings of the
pp. 3699–3708, 2008.
IEEE/ACM International Symposium on Cluster, Cloud and Grid Com-
[20] M. Shiraz and A. Gani, “A Lightweight Active Service Migration puting, 2016.
Framework for Computational Offloading in Mobile Cloud Computing,” [41] A. Iosup, S. Ostermann, N. Yigitbasi, R. Prodan, T. Fahringer, and
The Journal of Supercomputing, vol. 68, no. 2, pp. 978–995, 2014. D. Epema, “Performance Analysis of Cloud Computing Services for
[21] W. Li, Y. Zhao, S. Lu, and D. Chen, “Mechanisms and challenges on Many-Tasks Scientific Computing,” IEEE Transactions on Parallel and
Mobility-augmented Service Provisioning for Mobile Cloud Comput- Distributed Systems, vol. 22, no. 6, pp. 931–945, 2011.
ing,” IEEE Communications Magazine, vol. 53, no. 3, pp. 89–97, 2015. [42] B. Varghese, O. Akgun, I. Miguel, L. Thai, and A. Barker, “Cloud
[22] H. Hromic, D. Le Phuoc, M. Serrano, A. Antonic, I. P. Zarko, C. Hayes, Benchmarking for Performance,” in Proceedings of the IEEE Internati-
and S. Decker, “Real Time Analysis of Sensor Data for the Internet of noal Conference on Cloud Computing Technology and Science, 2014,
Things by Means of Clustering and Event Processing,” in Proceedings pp. 535–540.
of the IEEE International Conference on Communications, 2015, pp. [43] B. F. Cooper, A. Silberstein, E. Tam, R. Ramakrishnan, and R. Sears,
685–691. “Benchmarking Cloud Serving Systems with YCSB,” in Proceedings of
[23] C. Meurisch, A. Seeliger, B. Schmidt, I. Schweizer, F. Kaup, and the ACM Symposium on Cloud Computing, 2010, pp. 143–154.
M. Mühlhäuser, “Upgrading Wireless Home Routers for Enabling Large- [44] E. Deelman, K. Vahi, G. Juve, M. Rynge, S. Callaghan, P. J. Maechling,
scale Deployment of Cloudlets,” in Mobile Computing, Applications, and R. Mayani, W. Chen, R. Ferreira da Silva, M. Livny, and K. Wenger,
Services, 2015, pp. 12–29. “Pegasus: A Workflow Management System for Science Automation,”
[24] J. Povedano-Molina, J. M. Lopez-Vega, J. M. Lopez-Soler, A. Corradi, Future Generation Computer Systems, vol. 46, pp. 17–35, 2015.
and L. Foschini, “DARGOS: A Highly Adaptable and Scalable Monitor- [45] K. Wolstencroft, R. Haines, D. Fellows, A. R. Williams, D. Withers,
ing Architecture for Multi-tenant Clouds,” Future Generation Computer S. Owen, S. Soiland-Reyes, I. Dunlop, A. Nenadic, P. Fisher, J. Bhagat,
Systems, vol. 29, no. 8, pp. 2041–2056, 2013. F. Bacall, A. Hardisty, A. N. de la Hidalga, M. P. B. Vargas, S. Sufi, and
[25] S. A. D. Chaves, R. B. Uriarte, and C. B. Westphall, “Toward an C. A. Goble, “The Taverna Workflow Suite: Designing and Executing
Architecture for Monitoring Private Clouds,” IEEE Communications Workflows of Web Services on the Desktop, Web or in the Cloud,”
Magazine, vol. 49, no. 12, pp. 130–137, 2011. Nucleic Acids Research, vol. 41, pp. 557–561, 2013.
[26] J. Montes, A. Sánchez, B. Memishi, M. S. Pérez, and G. Antoniu, [46] J. Goecks, A. Nekrutenko, and J. Taylor, “Galaxy: A Comprehensive
“GMonE: A Complete Approach to Cloud Monitoring,” Future Gen- Approach for Supporting Accessible, Reproducible, and Transparent
eration Computer Systems, vol. 29, no. 8, pp. 2026–2040, 2013. Computational Research in the Life Sciences,” Genome Biology, vol. 11,
no. 8, pp. 1–13, 2010.
[27] J. S. Ward and A. Barker, “Varanus: In Situ Monitoring for Large Scale [47] J. Ruiz, J. Garrido, J. Santander-Vela, S. Snchez-Expsito, and L. Verdes-
Cloud Systems,” in Proceedings of the IEEE International Conference Montenegro, “AstroTaverna-Building Workflows with Virtual Observa-
on Cloud Computing Technology and Science, vol. 2, 2013, pp. 341–344. tory Services,” Astronomy and Computing, vol. 78, pp. 3 – 11, 2014.
[28] A. Li, X. Zong, S. Kandula, X. Yang, and M. Zhang, “CloudProphet: [48] S. Kartakis and J. A. McCann, “Real-time Edge Analytics for Cyber
Towards Application Performance Prediction in Cloud,” SIGCOMM Physical Systems Using Compression Rates,” in Proceedings of the
Computer Communication Review, vol. 41, no. 4, pp. 426–427, 2011. International Conference on Autonomic Computing, 2014, pp. 153–159.
[29] N. Grozev and R. Buyya, “Inter-Cloud Architectures and Application [49] I. Santos, M. Tilly, B. Chandramouli, and J. Goldstein, “DiAl: Dis-
Brokering: Taxonomy and Survey,” Software: Practice and Experience, tributed Streaming Analytics Anywhere, Anytime,” Proceedings of the
vol. 44, no. 3, pp. 369–390, 2014. VLDB Endowment, vol. 6, no. 12, pp. 1386–1389, 2013.
[30] S. K. Garg, S. Versteeg, and R. Buyya, “A Framework for Ranking [50] L. Xu, Z. Wang, and W. Chen, “The Study and Evaluation of ARM-
of Cloud Computing Services,” Future Generation Computer Systems, based Mobile Virtualization,” International Journal of Distributed Sen-
vol. 29, no. 4, pp. 1012–1023, 2013. sor Networks, Jan. 2015.
[31] A. J. Ferrer, F. HernáNdez, J. Tordsson, E. Elmroth, A. Ali-Eldin, [51] J. Andrus, C. Dall, A. V. Hof, O. Laadan, and J. Nieh, “Cells: A
C. Zsigri, R. Sirvent, J. Guitart, R. M. Badia, K. Djemame, W. Ziegler, Virtual Mobile Smartphone Architecture,” in Proceedings of the ACM
T. Dimitrakos, S. K. Nair, G. Kousiouris, K. Konstanteli, T. Varvarigou, Symposium on Operating Systems Principles, 2011, pp. 173–187.
B. Hudzia, A. Kipp, S. Wesner, M. Corrales, N. Forgó, T. Sharif, [52] M. Schwarzkopf, D. Murray, and S. Hand, “The Seven Deadly Sins of
and C. Sheridan, “OPTIMIS: A Holistic Approach to Cloud Service Cloud Computing Research,” in Proceedings of the USENIX Conference
Provisioning,” Future Generation Computer Systems, vol. 28, no. 1, pp. on Hot Topics in Cloud Computing, 2012.
66–77, 2012. [53] A. Barker, B. Varghese, J. S. Ward, and I. Sommerville, “Academic
[32] D. Yin and T. Kosar, “A Data-aware Workflow Scheduling Algorithm Cloud Computing Research: Five Pitfalls and Five Opportunities,”
for Heterogeneous Distributed Systems,” in International Conference on in Proceedings of the USENIX Conference on Hot Topics in Cloud
High Performance Computing and Simulation, 2011, pp. 114–120. Computing, 2014.
[33] T. Ghafariana and B. Javadi, “Cloud-aware Data Intensive Workflow
Scheduling on Volunteer Computing Systems,” Future Generation Com-
puter Systems, vol. 51, pp. 87–97, 2015.