bmd007 / fencex

Using kafkastreams as a platfrom for making serverless (faas) possible

Stateful Geofencing Faas

Abstract

To check if a coordinate (location, point on a map) is within a specific area is called geofencing. Since the emergence of smartphones, smart cars, and other mobile objects with GPS sensor attached to them, functional and non-functional requirements of geofencing systems have significantly changed. Previously, the number of mobile objects and thus the size of geospatial data about them and their load on the geofencing systems were much lower. Also, geofencing services were not much integrated into the day-to-day life of nonprofessional users. Nowadays, on the other hand, geofencing systems are required to process large loads of requests and data in realtime with a much higher availability while serving a wider range of users. Online taxi providers, elderly care platforms, pet tracking systems, and disaster management systems are examples of use cases in which geofencing with aforementioned non-functional requirements is in action. In this thesis we have designed and implemented a geofencing system called FenceX with those requirements in mind. The main pillars of FenceX, which allow for satisfaction of those requirements, are stream processing and microservices. FenceX has shown weak and strong scalability characteristics while being evaluated using real taxi trip data. Apart from being a geofencing system, FenceX can be considered an implementation for the idea of stateful function as a service (FaaS). Currently the available FaaS providers, Lambda from AWS most famously, do not allow for stateful functions. In this thesis, thanks to the usage of Kafka Stream as the stream processing framework and its supports for stateful operations, FenceX allows for the definition and deployment of stateful functions. While comparing FenceX with related works, we observed better results in multiple dimensions, peak throughput and resiliency for example.

Please read ./work-report/report-draft1.tex for more details

Technical requirement for starting development in this repo:

1- Due to limitations from spring-boot.experimental-thingJar, you have to use Java 15 during developemnt (for building project).

fencex's Issues

Microservices intercommunication

Sadi (March 28'th):
"Page 17: Figure 2.2: Physical data ow graph of FenceX: check the following text, this text you have to write it in a professional way as a scenario, please check section 3.2 architecture in this paper https://dl.acm.org/doi/abs/10.1145/3410992.3411003?casa_token=oPuFnFzRpc8AAAAA:8NpNT43DX17fqHKbqn8k770EtUFADZFECaVMsbAv78urchcGekUpibLSeTfv82Z9Rb0bedFSwBU"

Related work

Sadi (March 22nd):
"Regarding the related work: what I would suggest is to increase this part and reformat it again write just two- three pages as connected text without having the paper splits, don’t mention the number of paper in the related work here because you are not doing systemic study."

just use this thesis, dissertation or this study.

Sadi (March 22nd):
"I think you don’t have to write in this report, just use this thesis, dissertation or this study. "

Experiments

Sadi (March 22nd):
" I would like to aske to reconstruct the experiments (evaluation) chapter and you have to mention the experiment setting for each rounds and have them in a table, you have very good results but needs more description, in this chapter I missed the discussion (think about it)."

Poll/Push

Sadi (March 28):
"In the introduction, try to represent push and poll in a graphical way (image)"

WKT and others

Sadi (March 28'th):
"In page 10 “WKT, WKB, GeoJSON and GML. GeoJson is JSON friendly, GML is of XML nature and WKB is in bytes.” Please explain them and list their pros and cons points, Also at the end, put a table that shows the comparison between these techniques to give the reason for chosen WKT."

90% Completion

Mohammad (April 4'th):
"I'm not sure if at this stage my report should be 100 % complete. I will check this with Salman and Filip."

@sztoor Salman (April 4'th):
"You report need to be 90% (approx.) ready which means there will be no new experiments or major section changes will be done later. "

Sadi (April 4'th):
"and you didn’t address all the comments what you have done around 5% of my comments!!"
...
"Please try to address 90% of my comments before the defence as Salman said."

move Chapter 5 to top

Sadi (March 22nd):
"Chapter 5 I would suggest to place it before the experiment chapter and move the future work with the conclusion in one chapter."

Docker compose

Sadi (March 28'th):
"Figure 3.1: is it correct Docker Compose or should be docker container (please double check with Salman)"

Distirubted vs Central

Sadi (March 28'th):
"Page 27, “Distributed: Has multiple components. “ for me doesn’t make sense; the system can be centralized and have multiple components, re-explain it again."