Molecular dynamics (MD) is a cornerstone of scientific computing, but strong scaling often collapses at high parallelism because communication is bursty and highly sensitive to tail latency. MD advances by repeating a fixed timestep loop (one iteration of force computation and state update), and performance is largely determined by how quickly timesteps complete. A key reason is that each timestep contains short, synchronized communication phases, followed by a global dependency before the next timestep. Wafer-scale chips (WSCs) offer cycle-level latency and high on-chip bandwidth, yet their 2D mesh fabrics can still suffer burst-induced queue buildup; existing wavelet scheduling relies on a static stride that either over-injects (triggering credit backpressure) or over-throttles (wasting bandwidth) as conditions evolve.

Molecular Dynamics (MD) simulations with first-principles accuracy are widely applied in various fields, including materials science and molecular pharmacology.Current research focus on reducing the solution time of ab initio molecular dynamics (AIMD) from both

Recently, many Remote Direct Memory Access (RDMA) congestion control (CC) algorithms have been proposed to ensure the performance of high-speed Data Center Networks (DCNs). However, there is an inherent feedback delay in end-to-end CC, resulting in the inability to exert control over each flow within its first RTT. Upon many-to-one (Incast), the immediate high queue introduces significant latency to the short flows. In this paper, we propose SIM, a Sub-RTT-based Incast mitigation scheme to enhance congestion control. SIM offers two key functionalities: it enables the detection and notification of Incast within Sub-RTT, and it provides an adaptive adjustment algorithm that responds based on the severity of the Incast. Simulation results show that SIM fundamentally reduces the peak queue of Incast and shortens 99% FCT slowdown of short flows by up to 70% and 47% compared to the state-of-the-art congestion control schemes in DCNs, respectively.

Nowadays, Remote Direct Memory Access (RDMA) is gaining popularity in data centers for low CPU overhead, high throughput, and ultra-low latency. As one of the state-of-the-art RDMA Congestion Control (CC) mechanisms, HPCC leverages the In-band Network Telemetry (INT) features to achieve accurate control and significantly shortens the Flow Completion Time (FCT) for short flows. However, there exists redundant INT information increasing the processing latency at switches and affecting flows’ throughput. Besides, its end-to-end feedback mechanism is not timely enough to help senders cope well with bursty traffic, and there still exists a high probability of triggering Priority-based Flow Control (PFC) pauses under large-scale incast. In this paper, we propose a Congestion-Aware (CA) control mechanism called CACC, which attempts to push CC to the theoretical low INT overhead and PFC pause delay. CACC introduces two CA algorithms to quantize switch buffer and egress port congestion, separately, along with a fine-grained window size adjustment algorithm at the sender. Specifically, the buffer CA algorithm perceives large-scale congestion that may trigger PFC pauses and provides early feedback, significantly reducing the PFC pause delay. The egress port CA algorithm perceives the link state and selectively inserts useful INT data, achieving lower queue sizes and reducing the average overhead per packet from 42 bytes to 2 bits. In our evaluation, compared with HPCC, PINT, and Bolt, CACC shortens the average and tail FCT by up to 27% and 60.1%, respectively.

The distribution of traffic shows a characteristic of different flow sizes in Data Center Networks (DCNs), which requires diverse demands for data transmission. However, most existing congestion control schemes treat all the flows equivalently and have a consistent control logic, which cannot meet the diverse demands of applications. In this paper, we propose FACC, a flow-size-aware congestion control scheme. In FACC, we design a distinguished congestion control logic to assign the transmission demands of different kinds of flows in the network. To meet the diverse demands, FACC provides an adaptable congestion window (cwnd) adjustment by assigning customized weights with a well-designed flow-size-aware reward function. Simulation results show that FACC can reduce the average FCT and the 99- th percentile FCT slowdown of short flows by 35% and 23% compared to the state-of-the-art congestion control schemes in DCNs, respectively.

In recent years, Remote Direct Memory Access (RDMA) has been widely deployed in data centers to provide low-latency and high-bandwidth services. To ensure high performance in RDMA networks, congestion control manages queue depth on switches to minimize queueing delays. Although HPCC, the state-of-the-art scheme, can significantly reduce the flow completion time (FCT) of short flows, it still suffers from slow convergence and unfairness, which will affect the tail FCT of large flows. In this paper, we first analyze the causes of these defects and then propose an improved scheme called F2-HPCC, which introduces a self-adjusting additive increase algorithm to accelerate converging and a sliding window algorithm to improve fairness. In our evaluation, F2- HPCC achieves faster convergence and fairer allocations without sacrificing queue length and shortens the tail FCT of large flows by up to 33% under real data center workloads.