Francisco D. Igual
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Francisco D. Igual has a total of 162 co-authors. Here are the top co-authors by number of shared publications:
- Enrique S. Quintana Orti
Universitat Politécnica de Valencia, Valencia, Spain
59 shared publications - Luis Costero
Universidad Complutense de Madrid
20 shared publications - Sandra Catalán
Universitat Jaume I
20 shared publications - Katzalin Olcoz
Universidad Complutense de Madrid
17 shared publications Rafael Mayo
Depto. de Ingeniería y Ciencia de Computadores, Universität Jaume I, Castellón, Spain
17 shared publications- Rafael Rodríguez-Sánchez
Universidad Complutense de Madrid
14 shared publications
Latest Publications
- 8 September 2026
- Published by Fundación Muy Interesante in Fundación Muy Interesante
Una GPU, muchas tareas y una IA que decide: cómo el aprendizaje por refuerzo optimiza su planificación
Cuando varias tareas de cálculo tienen que compartir una misma tarjeta gráfica, alguien —o algo— tiene que decidir cómo repartírsela: si conviene trocearla en piezas pequeñas para muchas tareas ligeras, reservarla entera para una tarea exigente, o ir cambiando esa configuración sobre la marcha según van llegando nuevos trabajos. Las GPUs más recientes de los centros de datos permiten justo eso, dividirse en varias GPUs virtuales aisladas entre sí mediante una tecnología llamada Multi-Instance GPU (MIG), y hasta pueden reconfigurarse en caliente sin interrumpir lo que ya se está ejecutando. El problema es que decidir bien cómo y cuándo trocear el dispositivo es un rompecabezas de una complejidad enorme, del tipo que los informáticos llaman NP-duro, y los métodos que se usaban hasta ahora recurrían a simplificaciones que dejaban mucho rendimiento sobre la mesa. Un equipo de la Universidad Complutense de Madrid ha entrenado un agente de Aprendizaje por Refuerzo Profundo para que aprenda, por ensayo y error, a tomar esas decisiones sin necesidad de las simplificaciones habituales, considerando a la vez qué tamaño de instancia asignar a cada tarea, en qué orden ejecutarlas y cuándo merece la pena reconfigurar la tarjeta. Los resultados muestran mejoras consistentes frente a los planificadores más recientes, especialmente en las situaciones más difíciles de gestionar: aquellas en las que conviven tareas con necesidades muy distintas entre sí. Este artículo cuenta cómo funciona esa propuesta y qué la hace distinta de los enfoques anteriores.
- 8 September 2026
- Published by Fundación Muy Interesante in Fundación Muy Interesante
Una GPU, muchas tareas y una IA que decide: cómo el aprendizaje por refuerzo optimiza su planificación
Cuando varias tareas de cálculo tienen que compartir una misma tarjeta gráfica, alguien —o algo— tiene que decidir cómo repartírsela: si conviene trocearla en piezas pequeñas para muchas tareas ligeras, reservarla entera para una tarea exigente, o ir cambiando esa configuración sobre la marcha según van llegando nuevos trabajos. Las GPUs más recientes de los centros de datos permiten justo eso, dividirse en varias GPUs virtuales aisladas entre sí mediante una tecnología llamada Multi-Instance GPU (MIG), y hasta pueden reconfigurarse en caliente sin interrumpir lo que ya se está ejecutando. El problema es que decidir bien cómo y cuándo trocear el dispositivo es un rompecabezas de una complejidad enorme, del tipo que los informáticos llaman NP-duro, y los métodos que se usaban hasta ahora recurrían a simplificaciones que dejaban mucho rendimiento sobre la mesa. Un equipo de la Universidad Complutense de Madrid ha entrenado un agente de Aprendizaje por Refuerzo Profundo para que aprenda, por ensayo y error, a tomar esas decisiones sin necesidad de las simplificaciones habituales, considerando a la vez qué tamaño de instancia asignar a cada tarea, en qué orden ejecutarlas y cuándo merece la pena reconfigurar la tarjeta. Los resultados muestran mejoras consistentes frente a los planificadores más recientes, especialmente en las situaciones más difíciles de gestionar: aquellas en las que conviven tareas con necesidades muy distintas entre sí. Este artículo cuenta cómo funciona esa propuesta y qué la hace distinta de los enfoques anteriores.
- Open Access
- 26 February 2026
- Published by Springer Nature in International Journal of Networked and Distributed Computing
- Vol. 14 (1) , 11
QAFHE: A QoS-Aware Framework for Heterogeneous and Dynamic Edge-to-Cloud Kubernetes Deployments
- 26 February 2026
- Published by Springer Nature in International Journal of Networked and Distributed Computing
- Vol. 14 (1) , 11
Open AccessQAFHE: A QoS-Aware Framework for Heterogeneous and Dynamic Edge-to-Cloud Kubernetes Deployments
- 17 February 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (4) , 175
A comprehensive evaluation of spatial co-execution on GPUs using MPS and MIG technologies
- 17 February 2026
- Published by Springer Nature in The Journal of Supercomputing
- Vol. 82 (4) , 175
A comprehensive evaluation of spatial co-execution on GPUs using MPS and MIG technologies
- 24 November 2025
- Published by Springer Nature
- p. 431-443
Generation of Mixed-Precision Kernels for Quantized Transformer Encoders with Exo
- 24 November 2025
- Published by Springer Nature
- p. 431-443
Generation of Mixed-Precision Kernels for Quantized Transformer Encoders with Exo
- 24 November 2025
- Published by Springer Nature
- p. 534-547
Evaluation of RVV-Enabled COTS Platforms with Matrix Multiplication and Exo
- 24 November 2025
- Published by Springer Nature
- p. 534-547
Evaluation of RVV-Enabled COTS Platforms with Matrix Multiplication and Exo
- Open Access
- 11 November 2025
- Published by Springer Nature in Research Square
A comprehensive evaluation of spatial co-execution on GPUs using MPS and MIG technologies
To mitigate the increasingly common underutilization of computational resources in modern GPUs, spatial sharing methods enable multiple applications to use them simultaneously. This work presents a comprehensive evaluation of NVIDIA's primary technologies to achieve that goal: Multi-Process Service (MPS) and Multi-Instance GPU (MIG). Our findings reveal a crucial trade-off between MPS's flexibility and MIG's isolation, and reveals many key insights for improving the co-execution strategy according to jobs profiles. In the most favorable scenarios, MPS improves performance by up to 30% and reduces energy by about 20%, using its provisioning option to avoid resource monopolization. However, under memory contention, it suffers severe degradation, worsening performance by around 30%. Conversely, MIG's full hardware isolation resolves memory contention, leading to more consistent improvements, but these gains are tempered by higher overhead, and its rigid scheme can degrade performance in certain cases.
- 11 November 2025
- Published by Springer Nature in Research Square
Open AccessA comprehensive evaluation of spatial co-execution on GPUs using MPS and MIG technologies
To mitigate the increasingly common underutilization of computational resources in modern GPUs, spatial sharing methods enable multiple applications to use them simultaneously. This work presents a comprehensive evaluation of NVIDIA's primary technologies to achieve that goal: Multi-Process Service (MPS) and Multi-Instance GPU (MIG). Our findings reveal a crucial trade-off between MPS's flexibility and MIG's isolation, and reveals many key insights for improving the co-execution strategy according to jobs profiles. In the most favorable scenarios, MPS improves performance by up to 30% and reduces energy by about 20%, using its provisioning option to avoid resource monopolization. However, under memory contention, it suffers severe degradation, worsening performance by around 30%. Conversely, MIG's full hardware isolation resolves memory contention, leading to more consistent improvements, but these gains are tempered by higher overhead, and its rigid scheme can degrade performance in certain cases.
- Open Access
- 9 November 2025
- Published by Elsevier in Future Generation Computer Systems
- Vol. 177
The cambrian explosion of mixed-precision matrix multiplication for quantized deep learning inference
- 9 November 2025
- Published by Elsevier in Future Generation Computer Systems
- Vol. 177
Open AccessThe cambrian explosion of mixed-precision matrix multiplication for quantized deep learning inference
- Open Access
- 15 September 2025
- Published by Springer Nature in Cluster Computing
- Vol. 28 (12) , 1-21
Cache-aware optimization of matrix multiplication and matrix factorizations on multicore processors
This paper advocates for a careful customization of the special general matrix multiplication () kernels that are invoked from blocked routines for several relevant matrix factorizations in LAPACK, in order to improve their performance on modern multicore processors with hierarchical cache memories. To achieve this, we leverage a refined analytical model to dynamically tune the cache configuration parameters of for these kernels, taking into account the matrix operands’ dimensions, in order to improve cache occupation. In addition, toward the same goal, we accommodate a flexible development of architecture-specific micro-kernels for that allows us to select the option that, depending on the operands’ dimensions, ameliorates cache utilization. Our experiments for the LU and QR factorizations on two platforms, equipped with ARM (NVIDIA Carmel) and x86 (AMD EPYC) multi-core processors, demonstrate the benefits of this approach in terms of a better cache utilization and, in general, higher performance. Moreover, they also reveal the delicate balance between optimizing for multi-threaded parallelism versus cache usage as well as the positive effects of software prefetching.
- 15 September 2025
- Published by Springer Nature in Cluster Computing
- Vol. 28 (12) , 1-21
Open AccessCache-aware optimization of matrix multiplication and matrix factorizations on multicore processors
This paper advocates for a careful customization of the special general matrix multiplication () kernels that are invoked from blocked routines for several relevant matrix factorizations in LAPACK, in order to improve their performance on modern multicore processors with hierarchical cache memories. To achieve this, we leverage a refined analytical model to dynamically tune the cache configuration parameters of for these kernels, taking into account the matrix operands’ dimensions, in order to improve cache occupation. In addition, toward the same goal, we accommodate a flexible development of architecture-specific micro-kernels for that allows us to select the option that, depending on the operands’ dimensions, ameliorates cache utilization. Our experiments for the LU and QR factorizations on two platforms, equipped with ARM (NVIDIA Carmel) and x86 (AMD EPYC) multi-core processors, demonstrate the benefits of this approach in terms of a better cache utilization and, in general, higher performance. Moreover, they also reveal the delicate balance between optimizing for multi-threaded parallelism versus cache usage as well as the positive effects of software prefetching.
- Open Access
- 7 August 2025
- Published by Springer Nature in Research Square
QAFHE: A QoS-aware framework forheterogeneous and dynamic edge-to-cloud Kubernetes deployments
The introduction of the edge-to-cloud continuum paradigm introduces new chal-lenges due to the heterogeneity of computing entities in deployments. Thesechallenges primarily affect the ability to maintain appropriate levels of Qualityof Service (QoS) when assigning workloads and requests to nodes, particularlyconcerning response times, as well as application-specific metrics. Moreover, inscenarios where computing elements are inherently dynamic in terms of avail-ability, computing power, or latency, efficiently assigning workloads to the mostappropriate computing element becomes an even more significant challenge. Cur-rent generic orchestrators, like Kubernetes, have shown themselves to be effectivein homogeneous and static environments, where the usual QoS-unaware schedul-ing strategies focus mainly on load balancing, neglecting aspects such as reducinglatency or constraining application-level metrics. In this study, we reveal thatgeneric orchestrators like Kubernetes fall short when QoS-agnostic policies areapplied to heterogeneous and dynamic edge-to-cloud environments. We introduceQafhe, a novel framework that integrates effortlessly into Kubernetes via a ser-vice. This framework is designed with a set of QoS-aware scheduling policies toeffectively address the heterogeneity and dynamicity found in numerous edge-to-cloud setups. Our experiments, specifically involving the deployment of inferenceservers across diverse nodes, show considerable improvements in response timesacross various dynamic scenarios involving devices with varying capabilities, suchas multi-core CPUs and diverse GPU types.
- 7 August 2025
- Published by Springer Nature in Research Square
Open AccessQAFHE: A QoS-aware framework forheterogeneous and dynamic edge-to-cloud Kubernetes deployments
The introduction of the edge-to-cloud continuum paradigm introduces new chal-lenges due to the heterogeneity of computing entities in deployments. Thesechallenges primarily affect the ability to maintain appropriate levels of Qualityof Service (QoS) when assigning workloads and requests to nodes, particularlyconcerning response times, as well as application-specific metrics. Moreover, inscenarios where computing elements are inherently dynamic in terms of avail-ability, computing power, or latency, efficiently assigning workloads to the mostappropriate computing element becomes an even more significant challenge. Cur-rent generic orchestrators, like Kubernetes, have shown themselves to be effectivein homogeneous and static environments, where the usual QoS-unaware schedul-ing strategies focus mainly on load balancing, neglecting aspects such as reducinglatency or constraining application-level metrics. In this study, we reveal thatgeneric orchestrators like Kubernetes fall short when QoS-agnostic policies areapplied to heterogeneous and dynamic edge-to-cloud environments. We introduceQafhe, a novel framework that integrates effortlessly into Kubernetes via a ser-vice. This framework is designed with a set of QoS-aware scheduling policies toeffectively address the heterogeneity and dynamicity found in numerous edge-to-cloud setups. Our experiments, specifically involving the deployment of inferenceservers across diverse nodes, show considerable improvements in response timesacross various dynamic scenarios involving devices with varying capabilities, suchas multi-core CPUs and diverse GPU types.
- 10 June 2025
- Published by Springer Nature
- p. 298-301
Exploring Performance in RISC-V Vector Architecture: A Preliminary Study Integrating RISC-V and SYCL
- 10 June 2025
- Published by Springer Nature
- p. 298-301
Exploring Performance in RISC-V Vector Architecture: A Preliminary Study Integrating RISC-V and SYCL