Related papers: Optimizing the SSD Burst Buffer by Traffic Detecti…
Attention-based large language models (LLMs) have transformed modern AI applications, but the quadratic cost of self-attention imposes significant compute and memory overhead. Dynamic sparsity (DS) attention mitigates this, yet its hardware…
Burstable billing is widely adopted in practice, e.g., by colocation data center providers, to charge for their users, e.g., data centers, for data transferring. However, there is still a lack of research on what the best way is for a user…
NAND flash memory is ubiquitous in everyday life today because its capacity has continuously increased and cost has continuously decreased over decades. This positive growth is a result of two key trends: (1) effective process technology…
Iterative methods are commonly used approaches to solve large, sparse linear systems, which are fundamental operations for many modern scientific simulations. When the large-scale iterative methods are running with a large number of ranks…
This paper studies how RAID (redundant array of independent disks) could take full advantage of modern SSDs (solid-state drives) with built-in transparent compression. In current practice, RAID users are forced to choose a specific RAID…
Model checkpoints are critical Deep Learning (DL) artifacts that enable fault tolerance for training and downstream applications, such as inference. However, writing checkpoints to persistent storage, and other I/O aspects of DL training,…
One practical open problem is the development of a distributed algorithm that achieves near-optimal utility using only a finite (and small) buffer size for queues in a stochastic network. This paper studies utility maximization (or cost…
Today's clusters often have to divide resources among a diverse set of jobs. These jobs are heterogeneous both in execution time and in their rate of arrival. Execution time heterogeneity has lead to the development of hybrid schedulers…
The development of high-speed storage devices such as NVMe SSDs has shifted the primary I/O bottleneck from hardware to software. Modern database systems also rely on kernel-based I/O paths, where frequent system call invocations and…
In recent years, numerous ideas have emerged for designing a mutually reinforcing mechanism or extra stages for the image fusion task, ignoring the inevitable gaps between different vision tasks and the computational burden. We argue that…
High Speed Downlink Packet Access (HSDPA) was introduced to UMTS radio access segment to provide higher capacity for new packet switched services. As a result, packet switched sessions with multiple diverse traffic flows such as concurrent…
Spiking neural networks (SNNs) promise energy-efficient computation by mimicking biological neural dynamics, yet existing plasticity rules focus on isolated spike pairs and fail to leverage the synchronous activity patterns that drive…
Stream analytics have an insatiable demand for memory and performance. Emerging hybrid memories combine commodity DDR4 DRAM with 3D-stacked High Bandwidth Memory (HBM) DRAM to meet such demands. However, achieving this promise is…
Parallelization of A* path planning is mostly limited by the number of possible motions, which is far less than the level of parallelism that modern processors support. In this paper, we go beyond the limitations of traditional parallelism…
Diffusion model deployment has been suffering from high energy consumption and inference latency despite its superior performance in visual generation tasks. Dynamic voltage and frequency scaling (DVFS) offers a promising solution to…
NAND flash memory is ubiquitous in everyday life today because its capacity has continuously increased and cost has continuously decreased over decades. This positive growth is a result of two key trends: (1) effective process technology…
Data are being collected from various aspects of life. These data can often arrive in chunks/batches. Traditional static clustering algorithms are not suitable for dynamic datasets, i.e., when data arrive in streams of chunks/batches. If we…
Pushing and caching hold the promise of significantly increasing the throughput of content-centric wireless networks. However, the throughput gain of these techniques is limited by the buffer size of the receiver. To overcome this, this…
Diffusion-based generation is increasingly powering production content pipelines; however, deploying these models at scale remains a significant challenge. Model weights frequently exceed the memory capacity of commodity GPUs, while the…
High intensive computation applications can usually take days to months to finish an execution. During this time, it is common to have variations of the available resources when considering that such hardware is usually shared among a…