Warning: pg_query(): Query failed: ERROR: missing chunk number 0 for toast value 29512337 in pg_toast_2619 in /dati/webiit-old/includes/database.pgsql.inc on line 138 Warning: ERROR: missing chunk number 0 for toast value 29512337 in pg_toast_2619 query: SELECT data, created, headers, expire, serialized FROM cache_page WHERE cid = 'https://www-old.iit.cnr.it/en/node/59638' in /dati/webiit-old/includes/database.pgsql.inc on line 159 Warning: pg_query(): Query failed: ERROR: missing chunk number 0 for toast value 29512337 in pg_toast_2619 in /dati/webiit-old/includes/database.pgsql.inc on line 138 Warning: ERROR: missing chunk number 0 for toast value 29512337 in pg_toast_2619 query: SELECT data, created, headers, expire, serialized FROM cache_page WHERE cid = 'https://www-old.iit.cnr.it/en/node/59638' in /dati/webiit-old/includes/database.pgsql.inc on line 159 Optimising cost vs accuracy of decentralised analytics in fog computing environments | IIT - CNR - Istituto di Informatica e Telematica
IIT Home Page CNR Home Page

Optimising cost vs accuracy of decentralised analytics in fog computing environments

The exponential growth of devices and data at theedges of the Internet is rising scalability and privacy concernson approaches based exclusively on remote cloud platforms. Datagravity, a fundamental concept in Fog Computing, points towardsdecentralisation of computation for data analysis, as a viablealternative to address those concerns. Decentralising AI taskson several cooperative devices means identifying the optimal setof locations or Collection Points (CP for short) to use, in thecontinuum between full centralisation (i.e., all data on a singledevice) and full decentralisation (i.e., data on source locations).We propose an analytical framework able to find the optimaloperating point in this continuum, linking the accuracy of thelearning task with the corresponding network and computationalcost for moving data and running the distributed training at theCPs. We show through simulations that the model accuratelypredicts the optimal trade-off, quite often an intermediate pointbetween full centralisation and full decentralisation, showing alsoa significant cost saving w.r.t. both of them. Finally, the analyticalmodel admits closed-form or numeric solutions, making it notonly a performance evaluation instrument but also a design toolto configure a given distributed learning task optimally beforeits deployment.


2020

IIT authors:

Type: Rapporto Tecnico
Field of reference: Computer Science & Engineering
IIT TR-22/2020

File: IIT-22-2020.pdf

Activity: Future Internet