arXiv subjects

Pedro Javier Garcia

Publications and source records attributed to Pedro Javier Garcia.

7 recordsLinked to original sources

arXiv2025

Improving dynamic congestion isolation in data-center networks

The rise of distributed AI and large-scale applications has impacted the communication operations of data-center and Supercomputer interconnection networks, leading to dramatic incast or in-network congestion scenarios and challenging existing congestion control mechanisms, such as injection throttling (e.g., DCQCN) or congestion isolation (CI). While DCQCN provides a scalable traffic rate adjustment for congesting flows at end nodes (which is slow) and CI effectively isolates these flows in special network resources (which requires extra logic in the switches), their combined use, although it diminishes their particular drawbacks, leads to false congestion scenarios identification and signaling, excessive throttling, and inefficient network resource utilization. In this paper, we propose a new CI mechanism, called Improved Congestion Isolation (ICI), which efficiently combines CI and DCQCN so that the information of the isolated congesting flows is used to guide the ECN marking performed by DCQCN in a way that victim flows do not end up being marked. This coordination reduces false-positive congestion detection, suppresses unnecessary closed-loop feedback (i.e., wrong congestion notifications), and improves responsiveness to communication microbursts. Evaluated under diverse traffic patterns, including incast and Data-center workloads, ICI reduces the number of generated BECNs by up to 32x and improves tail latency by up to 31%, while maintaining high throughput and scalability.

cs.NI↗

arXiv2025

Scalable and Efficient Intra- and Inter-node Interconnection Networks for Post-Exascale Supercomputers and Data centers

The rapid growth of data-intensive applications such as generative AI, scientific simulations, and large-scale analytics is driving modern supercomputers and data centers toward increasingly heterogeneous and tightly integrated architectures. These systems combine powerful CPUs and accelerators with emerging high-bandwidth memory and storage technologies to reduce data movement and improve computational efficiency. However, as the number of accelerators per node increases, communication bottlenecks emerge both within and between nodes, particularly when network resources are shared among heterogeneous components.

cs.AR↗

arXiv2025

On the Impact of Intra-node Communication in the Performance of Supercomputer and Data Center Interconnection Networks

In the last decade, specific-purpose computing and storage devices, such as GPUs, TPUs, or high-speed storage, have been incorporated into server nodes of Supercomputers and Data centers. The development of high-bandwidth memory (HBM) enabled a much more compact form factor for these devices, thus allowing the interconnection of several of them within a server node, typically using an intra-node interconnection network (e.g., PCIe, NVLink, or Infinity Fabric). These networks allow scaling up the number of specific computing and storage devices per node. Furthermore, the inter-node networks communicate thousands of these devices placed in different server nodes in a Supercomputer or Data Center. Unfortunately, the intra- and inter-node networks may become the system's bottleneck due to the increasing communication demand among accelerators of applications such as generative AI. Although current intra-node network designs alleviate this bottleneck by increasing the bandwidth of the intra-node network, we show in this paper that such a high bandwidth for intra-node communication may hinder the inter-node communication performance when traffic from outside the node arrives at the intra-node devices, resulting in interference with intra-node traffic. To analyze the impact of this interference, we have studied the communication operations of realistic traffic patterns exploiting intra-node communication. We have developed a generic intra- and inter-node simulation model based on OMNeT++ and modeled the mentioned communication operations. We have also performed extensive simulation experiments that confirm that increasing the intra-node network bandwidth and the number of computing devices per node (i.e., accelerators) is counterproductive to the inter-node communication performance.

cs.AR↗

arXiv2025

Leveraging InfiniBand Controller to Configure Deadlock-Free Routing Engines for Dragonflies

The Dragonfly topology is currently one of the most popular network topologies in high-performance parallel systems. The interconnection networks of many of these systems are built from components based on the InfiniBand specification. However, due to some constraints in this specification, the available versions of the InfiniBand network controller (OpenSM) do not include routing engines based on some popular deadlock-free routing algorithms proposed theoretically for Dragonflies, such as the one proposed by Kim and Dally based on Virtual-Channel shifting. In this paper we propose a straightforward method to integrate this routing algorithm in OpenSM as a routing engine, explaining in detail the configuration required to support it. We also provide experiment results, obtained both from a real InfiniBand-based cluster and from simulation, to validate the new routing engine and to compare its performance and requirements against other routing engines currently available in OpenSM.

cs.NI↗

arXiv2022

High-Quality Fault Resiliency in Fat Trees

Coupling regular topologies with optimised routing algorithms is key in pushing the performance of interconnection networks of supercomputers.In this paper we present Dmodc, a fast deterministic routing algorithm for Parallel Generalised Fat-Trees (PGFTs) which minimises congestion risk even under massive network degradation caused by equipment failure.Dmodc computes forwarding tables with a closed-form arithmetic formula by relying on a fast preprocessing phase.This allows complete re-routing of networks with tens of thousands of nodes in less than a second.In turn, this greatly helps centralised fabric management react to faults with high-quality routing tables and no impact to running applications in current and future very large-scale HPC clusters.

cs.DC↗

arXiv2022

Node-Type-Based Load-Balancing Routing for Parallel Generalized Fat-Trees

High-Performance Computing (HPC) clusters are made up of a variety of node types (usually compute, I/O, service, and GPGPU nodes) and applications don't use nodes of a different type the same way. Resulting communication patterns reflect organization of groups of nodes, and current optimal routing algorithms for all-to-all patterns will not always maximize performance for group-specific communications. Since application communication patterns are rarely available beforehand, we choose to rely on node types as a good guess for node usage. We provide a description of node type heterogeneity and analyse performance degradation caused by unlucky repartition of nodes of the same type. We provide an extension to routing algorithms for Parallel Generalized Fat-Tree topologies (PGFTs) which balances load amongst groups of nodes of the same type. We show how it removes these performance issues by comparing results in a variety of situations against corresponding classical algorithms.

cs.DC↗

arXiv2022

High-Quality Fault-Resiliency in Fat-Tree Networks (Extended Abstract)

Coupling regular topologies with optimized routing algorithms is key in pushing the performance of interconnection networks of HPC systems. In this paper we present Dmodc, a fast deterministic routing algorithm for Parallel Generalized Fat-Trees (PGFTs) which minimizes congestion risk even under massive topology degradation caused by equipment failure. It applies a modulo-based computation of forwarding tables among switches closer to the destination, using only knowledge of subtrees for pre-modulo division. Dmodc allows complete rerouting of topologies with tens of thousands of nodes in less than a second, which greatly helps centralized fabric management react to faults with high-quality routing tables and no impact to running applications in current and future very large-scale HPC clusters. We compare Dmodc against routing algorithms available in the InfiniBand control software (OpenSM) first for routing execution time to show feasibility at scale, and then for congestion risk under degradation to demonstrate robustness. The latter comparison is done using static analysis of routing tables under random permutation (RP), shift permutation (SP) and all-to-all (A2A) traffic patterns. Results for Dmodc show A2A and RP congestion risks similar under heavy degradation as the most stable algorithms compared, and near-optimal SP congestion risk up to 1% of random degradation.

cs.DC↗