David Carrera

Research

Fifteen years on the performance management of data centre workloads, from application servers to the edge.

Background

My research was about the performance management of data centre workloads: deciding where work runs, and when, so that shared infrastructure meets its goals. It began with my PhD on heterogeneous workload management in clouds, moved to the autonomic placement of mixed batch and transactional workloads with IBM Research, and to deadline-aware scheduling for MapReduce.

At the Barcelona Supercomputing Center I led the Data-Centric Computing research group, and with it we widened that work to big data, IoT streams, GPUs and non-volatile memory, and to AI for resource management: learning models, including conditional restricted Boltzmann machines, that predict how workloads behave and steer data centre operations. With industrial partners including IBM, Microsoft, Intel and Cisco, the same question reached fog and edge computing, where infrastructure is distributed and heterogeneous. Nearby Computing grew out of that last step.

Themes

  • AI for resource management

    Machine learning models of workloads, from classical methods to deep learning and conditional restricted Boltzmann machines, to predict demand and guide placement and data centre operations.

  • Workload placement and scheduling

    Holistic optimisation of software-defined data centres, with performance models that span heterogeneous infrastructure and workloads.

  • Fog and edge computing

    Bridging cloud and edge for NFV and 5G, and the architectures that let cities and networks run applications close to where data is produced.

  • Big data cost-effectiveness

    How configuration choices drive the runtime and price of Hadoop and Spark deployments. The group built the ALOJA open benchmarking platform.

  • IoT stream processing

    Real-time composition, transformation and filtering of data streams. The group built the servIoTicy platform.

  • Storage and acceleration

    Non-volatile memory, GPUs and FPGAs for data-centric and IO-bound applications.

Selected publications

AI and GPU management, service placement

  • 2022Burst-aware predictive autoscaling for containerized microservicesM. Abdullah, W. Iqbal, J. Ll. Berral, J. Polo, D. Carrera. IEEE Transactions on Services Computing 15(3). doi:10.1109/tsc.2020.2995937
  • 2017Topology-aware GPU scheduling for learning workloads in cloud environmentsM. Amaral, J. Polo, D. Carrera, S. Seelam, M. Steinder. SC17, International Conference for High Performance Computing, Networking, Storage and Analysis. doi:10.1145/3126908.3126933
  • 2024Dexter: a performance-cost efficient resource allocation manager for serverless data analyticsA. M. Nestorov, D. Marrón, A. Gutierrez-Torre, C. Wang, C. Misale, A. Youssef, D. Carrera, J. Ll. Berral. 25th International Middleware Conference. doi:10.1145/3652892.3700753

AI for data centre operations

  • 2020Adaptive sliding windows for improved estimation of data center resource utilizationW. Iqbal, J. Ll. Berral, D. Carrera. Future Generation Computer Systems 104. doi:10.1016/j.future.2019.10.026
  • 2019Adaptive prediction models for data center resources utilization estimationW. Iqbal, J. Ll. Berral, A. Erradi, D. Carrera. IEEE Transactions on Network and Service Management 16(4). doi:10.1109/tnsm.2019.2932840
  • 2020A highly parameterizable framework for Conditional Restricted Boltzmann Machine based workloads accelerated with FPGAs and OpenCLZ. Jaksic, N. Cadenelli, D. Buchaca, J. Polo, J. Ll. Berral, D. Carrera. Future Generation Computer Systems 104. doi:10.1016/j.future.2019.10.025
  • 2018Automatic generation of workload profiles using unsupervised learning pipelinesD. Buchaca, J. Ll. Berral, D. Carrera. IEEE Transactions on Network and Service Management 15(1). doi:10.1109/tnsm.2017.2786047

Edge and fog computing

  • 2022Autonomous lifecycle management for resource-efficient workload orchestration for green edge computingF. Guim, T. Metsch, H. Moustafa, T. Verrall, D. Carrera, N. Cadenelli, et al. IEEE Transactions on Green Communications and Networking 6(1). doi:10.1109/tgcn.2021.3127531
  • 2017A new era for cities with fog computingM. Yannuzzi, F. van Lingen, A. Jain, O. Lluch, M. Mendoza, D. Carrera, et al. IEEE Internet Computing 21(2). doi:10.1109/mic.2017.25
  • 2017The unavoidable convergence of NFV, 5G, and fog: a model-driven approach to bridge cloud and edgeF. van Lingen, M. Yannuzzi, A. Jain, R. Irons-Mclean, O. Lluch, D. Carrera, et al. IEEE Communications Magazine 55(8). doi:10.1109/mcom.2017.1600907
  • 2022Automatic distributed deep learning using resource-constrained edge devicesA. Gutierrez-Torre, K. Bahadori, W. Iqbal, T. Vardanega, J. Ll. Berral, D. Carrera. IEEE Internet of Things Journal 9(16). doi:10.1109/jiot.2021.3098973

Big data

  • 2010Performance-driven task co-scheduling for MapReduce environmentsJ. Polo, D. Carrera, Y. Becerra, J. Torres, E. Ayguadé, M. Steinder, I. Whalley. IEEE/IFIP Network Operations and Management Symposium (NOMS). doi:10.1109/noms.2010.5488494
  • 2011Resource-aware adaptive scheduling for MapReduce clustersJ. Polo, C. Castillo, D. Carrera, Y. Becerra, I. Whalley, M. Steinder, J. Torres, E. Ayguadé. ACM/IFIP/USENIX Middleware. doi:10.1007/978-3-642-25821-3_10
  • 2014ALOJA: a systematic study of Hadoop deployment variables to enable automated characterization of cost-effectivenessN. Poggi, D. Carrera, A. Call, S. Mendoza, Y. Becerra, J. Torres, et al. IEEE International Conference on Big Data. doi:10.1109/bigdata.2014.7004322

More than 100 peer-reviewed papers in all. The full list is on Google Scholar and DBLP.

Research profiles