Paper accepted to ICDE 2024

COSTREAM: Learned Cost Models for Operator Placement in Edge-Cloud Environments


Authors: Roman Heinrich, Carsten Binnig, Harald Kornmayer, Manisha Luthra

We are happy to announce that our paper COSTREAM was accepted to the International Conference of Data Engineering (ICDE) 2024, Utrecht, Netherlands.

We propose COSTREAM, a learned zero-shot cost model that predicts execution costs for streaming queries within edge-cloud distributed environments. The cost model aids in the initial operator placement of operators across diverse and unseen hardware. The model has been evaluated to provide highly accurate estimates and can facilitate a median speed-up of 21× over baseline methods when optimizing operator placements.

It can be downloaded here.