A declarative solution for placing and configuring applications in Osmotic Computing settings.

Osmolog methodology is described and assessed in:

Stefano Forti, Antonio Brogi
Declarative Osmotic Application Placement,
Next Generation Information Systems (NeGIS) @ CAiSE, 2021.

If you wish to reuse source code in this repo, please consider citing the above mentioned article.

Background and Requirements

osmolog is written in Prolog. Prolog programs are finite sets of clauses of the form:

a :- b1, ... , bn.

stating that a holds when b1 and ... and bn holds, where n >= 0 and a, b1 ..., bn are atomic literals. Clauses with empty condition are also called facts. Prolog variables begin with upper-case letters, lists are denoted by square brackets, and negation by \+.

To run osmolog, please install SWI-Prolog.

Quickstart Example

Consider the Osmotic application below from Augmented Reality.

It is made of four MicroELements (MELs), some of which exist in more than one version with different IoT, software and hardware requirements. Versions range from the less demanding light version (i.e. triangles) to a medium version (i.e. squares) to a full version (i.e. circles). Those three versions (or flavours) are suited for IoT, Edge and Cloud devices respectively.

Given a Cloud-IoT infrastructure, osmolog jointly determines a solution to these placement-related questions:

Where to deploy each MEL composing the application?

and

Which MEL version to deploy?

Model

We first describe the osmolog model, following the input data contained in the file example.pl.

Application

A fully adaptive version of the application above can be specified as in:

mel((usersData,full), [docker], 64, []).

mel((videoStorage,full), [docker], 16, []).
mel((videoStorage,medium), [docker], 8, []).

mel((movementProcessing,full), [docker], 8, []).
mel((movementProcessing,medium), [gcc, make], 4, []).

mel((arDriver,full), [docker], 4, [phone, lightSensor]).
mel((arDriver,medium), [gcc,caffe], 2, [phone, lightSensor]).
mel((arDriver,light), [gcc], 1, [phone]).

mel2mel(usersData, videoStorage, 70).
mel2mel(videoStorage, movementProcessing, 30).
mel2mel(movementProcessing, arDriver, 20).

application((arApp, adaptive), [(usersData,full), (videoStorage,_), (movementProcessing,_), (arDriver,_)]).

Note that mel/4 facts denote all MEL requirements for different versions (MelId, Version) in terms of software requirements, hardware resources and IoT requirements. Besides, mel2mel/3 denote latency requirements in milliseconds between application MELs. Finally, application/2 facts denote instead the services composing a certain version (AppId, Version) of the considered application.

Infrastructure

A Cloud-IoT infrastructure of two nodes is declared as in

node(edge42, [(gcc,0),(caffe,4)], (6, 3), [(phone,1),(lightSensor,1)]).
node(cloud42, [(docker, 5)], (100, 1), []).

link(edge42, cloud42, 20).

Note that node/4 facts denote the software, hardware and IoT capabilities of each node, associated with their estimated monthly usage cost. Finally, link/3 facts denote the end-to-end latency in milliseconds between two nodes.

Osmolog at Work

Osmolog determines eligible placements (i.e. mapping from MELs to Cloud-IoT nodes) and configuration (i.e. mapping to MELs from one of their versions) of a version of an Osmotic application. Placements are ranked so to:

• minimise estimated operational costs, and
• maximise compliance to a preferred MEL version among light, medium and full.

A (best) candidate placement can be determined by means of an exhaustive (exhaustive.pl) or a heuristic search strategy (greedy.pl).

Exhaustive Search

To use the exhaustive search (exhaustive.pl) to determine a placement for the example application onto the example infrastructure, simply query the predicate:

% goForBest(SortType, AppId, AppVersion, PreferredMELVersion, MaxCost, BestPlacement).
?- goForBest((0,highest), arApp, adaptive, full, 110, Best).

This query returns the best possible placement BestPlacement of the arApp, maximising the number of full MELs and minimising the operational cost, without exceeding 110 euro per month.

The output BestPlacement is

BestPlacement = [175, 75, 107, [on(usersData, full, cloud42), on(videoStorage, full, cloud42), on(movementProcessing, full, cloud42), on(arDriver, light, edge42)]] 

ranked 175/200, and featuring 75% version compliance and an estimated monthly cost of 107 euro.

Heuristic Search

To use the exhaustive search (greedy.pl) to determine a placement for the example application onto the example infrastructure, simply query the predicate:

% h_placement(Application, V, PrefVersion, CapCost, HPlacement, VersionCompliance, Cost).
?- h_placement(arApp, adaptive, full, 110, HPlacement, VC, C).

This query returns a (sub-)optimal HPlacement of the arApp, trying to maximise the number of full MELs and trying to minimise the operational cost, without exceeding 110 euro per month.

The output HPlacement is the same as before

HPlacement = [on(arDriver, light, edge42), on(videoStorage, full, cloud42), on(movementProcessing, full, cloud42), on(usersData, full, cloud42)],
VC = 75,
C = 107.

featuring 75% version compliance and an estimated monthly cost of 107 euro. Note that, from our experiments, the heuristic search determines solutions that are on average 9% far from the optimal, achieving a 40x speed-up on execution times.