Search

Fault detection and localization are vital for ensuring the stability of data center networks (DCNs). Specifically, adaptive fault diagnosis is deemed a fundamental technology in achieving the fault tolerance of systems. The highly scalable data center network (HSDC) is a promising structure of server-centric DCNs, as it exhibits the capacity for incremental scalability, coupled with the assurance of low cost and energy consumption, low diameter, and high bisection width. In this paper, we first determine that both the connectivity and diagnosability of the m-dimensional complete HSDC, denoted by H S D C m (m) , are m. Further, we propose an efficient adaptive fault diagnosis algorithm to diagnose an H S D C m (m) within three test rounds, and at most N + 4 m (m − 2) tests with m ≥ 3 (resp. at most nine tests with m = 2 ), where N = m · 2 m is the total number of nodes in H S D C m (m) . Our experimental outcomes demonstrate that this diagnosis scheme of HSDC can achieve complete diagnosis and significantly reduce the number of required tests. [ABSTRACT FROM AUTHOR]

"Linearly many faults" is a phenomenon observed by Cheng and Lipták in which a specific structure emerges when a graph is disconnected and often occurs in various interconnection networks. This phenomenon means that if a certain number of vertices or edges are deleted from a graph, the remaining part either stays connected or breaks into one large component along with smaller components with just a few vertices. This phenomenon can be observed in many types of graphs and has important implications for network analysis and optimization. In this paper, we first validate the phenomenon of linearly many faults for surviving graph of a burnt pancake graph B P n when removing any edge subset with a size of approximately six times λ (B P n) . For graph G, the ℓ-component edge connectivity denoted as λ ℓ (G) (resp., the ℓ-extra edge connectivity denoted as λ (ℓ) (G) ) is the cardinality of a minimum edge subset S such that G − S is disconnected and has at least ℓ components (resp., each component of G − S has at least ℓ + 1 vertices). Both λ ℓ (G) and e λ (ℓ) (G) are essential metrics for network reliability assessment. Specifically, from the property of "linearly many faults", we may further prove that λ 5 (B P n) = λ (3) (B P n) + 3 = 4 n − 3 for n ⩾ 5 ; λ 6 (B P n) = λ (4) (B P n) + 4 = 5 n − 4 and λ 7 (B P n) = λ (5) (B P n) + 5 = 6 n − 5 for n ⩾ 6 . [ABSTRACT FROM AUTHOR]

Traditional model compression techniques are dependent on handcrafted features and require domain experts, with a tradeoff between model size, speed, and accuracy. This study proposes a new approach toward resolving model compression problems. Our approach combines reinforcement-learning-based automated pruning and knowledge distillation to improve the pruning of unimportant network layers and the efficiency of the compression process. We introduce a new state quantity that controls the size of the reward and an attention mechanism that reinforces useful features and attenuates useless features to enhance the effects of other features. The experimental results show that the proposed model is superior to other advanced pruning methods in terms of the computation time and accuracy on CIFAR-100 and ImageNet dataset, where the accuracy is approximately 3% higher than that of similar methods with shorter computation times. [ABSTRACT FROM AUTHOR]

The design of interconnection networks is a fundamental aspect of high-performance computing (HPC) systems. Among the available topologies, the Galaxyfly network stands out as a low-diameter and flexible-radix network for HPC applications. Given the paramount importance of collective communication in HPC performance, in this paper, we present two different all-to-all broadcast algorithms for the Galaxyfly network, which adhere to the supernode-first rule and the router-first rule, respectively. Our performance evaluation validates their effectiveness and shows that the first algorithm has a higher degree of utilization of network channels, and that the second algorithm can significantly reduce the average time for routers to collect packets from the supernode. [ABSTRACT FROM AUTHOR]