arXiv subjects

Hoang Nam Nguyen

Publications and source records attributed to Hoang Nam Nguyen.

3 recordsLinked to original sources

arXiv2026

HCCL: Collective Communication for Meta Training and Inference Accelerators

We present HCCL, a collective communication library co-designed with Meta's MTIA 300 accelerator, the first Meta chip to integrate backend networking directly on chip package. MTIA 300 includes dedicated message engines (MEs) with near-memory compute (NMC) that fully offload collective execution from the compute grid, enabling large overlap between computation and communication. HCCL uses a compiled communication model in which the host generates a complete description of each collective including dependencies. We describe the control and data path architecture, topology-aware algorithm selection across MTIA 300's asymmetric scale-up and scale-out network, and optimizations for both training and inference workloads. For training, HCCL achieves up to 940 GB/s on intra-rack collectives while introducing less than 0.5% degradation to concurrent compute throughput. For inference, we leverage one-sided communication primitives that bypass the scheduling path to minimize collective latency and describe collective designs that improve compute-communication pipelining for latency-sensitive workloads.

cs.NI↗

arXiv2025

Deadline-Aware Joint Task Scheduling and Offloading in Mobile Edge Computing Systems

The demand for stringent interactive quality-of-service has intensified in both mobile edge computing (MEC) and cloud systems, driven by the imperative to improve user experiences. As a result, the processing of computation-intensive tasks in these systems necessitates adherence to specific deadlines or achieving extremely low latency. To optimize task scheduling performance, existing research has mainly focused on reducing the number of late jobs whose deadlines are not met. However, the primary challenge with these methods lies in the total search time and scheduling efficiency. In this paper, we present the optimal job scheduling algorithm designed to determine the optimal task order for a given set of tasks. In addition, users are enabled to make informed decisions for offloading tasks based on the information provided by servers. The details of performance analysis are provided to show its optimality and low complexity with the linearithmic time O(nlogn), where $n$ is the number of tasks. To tackle the uncertainty of the randomly arriving tasks, we further develop an online approach with fast outage detection that achieves rapid acceptance times with time complexity of O(n). Extensive numerical results are provided to demonstrate the effectiveness of the proposed algorithm in terms of the service ratio and scheduling cost.

cs.DC↗

arXiv2022

Distributionally robust chance-constrained Markov decision processes

Markov decision process (MDP) is a decision making framework where a decision maker is interested in maximizing the expected discounted value of a stream of rewards received at future stages at various states which are visited according to a controlled Markov chain. Many algorithms including linear programming methods are available in the literature to compute an optimal policy when the rewards and transition probabilities are deterministic. In this paper, we consider an MDP problem where the transition probabilities are known and the reward vector is a random vector whose distribution is partially known. We formulate the MDP problem using distributionally robust chance-constrained optimization framework under various types of moments based uncertainty sets, and statistical-distance based uncertainty sets defined using phi-divergence and Wasserstein distance metric. For each type of uncertainty set, we consider the case where a random reward vector has either a full support or a nonnegative support. For the case of full support, we show that the distributionally robust chance-constrained Markov decision process is equivalent to a second-order cone programming problem for the moments and phi-divergence distance based uncertainty sets, and it is equivalent to a mixed-integer second-order cone programming problem for an Wasserstein distance based uncertainty set. For the case of nonnegative support, it is equivalent to a copositive optimization problem and a biconvex optimization problem for the moments based uncertainty sets and Wasserstein distance based uncertainty set, respectively. As an application, we study a machine replacement problem and illustrate numerical experiments on randomly generated instances.

math.OC↗