Guoqing Harry Xu

John Thorpe, Pengzhan Zhao, Jonathan Eyolfson et al.

DNN models across many domains continue to grow in size, resulting in high resource requirements for effective training, and unpalatable (and often unaffordable) costs for organizations and research labs across scales. This paper aims to significantly reduce training costs with effective use of preemptible instances, i.e., those that can be obtained at a much cheaper price while idle, but may be preempted whenever requested by priority users. Doing so, however, requires new forms of resiliency and efficiency to cope with the possibility of frequent preemptions - a failure model that is drastically different from the occasional failures in normal cluster settings that existing checkpointing techniques target. We present Bamboo, a distributed system that tackles these challenges by introducing redundant computations into the training pipeline, i.e., whereby one node performs computations over not only its own layers but also over some layers in its neighbor. Our key insight is that training large models often requires pipeline parallelism where "pipeline bubbles" naturally exist. Bamboo carefully fills redundant computations into these bubbles, providing resilience at a low cost. Across a variety of widely used DNN models, Bamboo outperforms traditional checkpointing by 3.7x in training throughput, and reduces costs by 2.4x compared to a setting where on-demand instances are used.

Arthi Padmanabhan, Neil Agarwal, Anand Iyer et al.

Video analytics pipelines have steadily shifted to edge deployments to reduce bandwidth overheads and privacy violations, but in doing so, face an ever-growing resource tension. Most notably, edge-box GPUs lack the memory needed to concurrently house the growing number of (increasingly complex) models for real-time inference. Unfortunately, existing solutions that rely on time/space sharing of GPU resources are insufficient as the required swapping delays result in unacceptable frame drops and accuracy violations. We present model merging, a new memory management technique that exploits architectural similarities between edge vision models by judiciously sharing their layers (including weights) to reduce workload memory costs and swapping delays. Our system, GEMEL, efficiently integrates merging into existing pipelines by (1) leveraging several guiding observations about per-model memory usage and inter-layer dependencies to quickly identify fruitful and accuracy-preserving merging configurations, and (2) altering edge inference schedules to maximize merging benefits. Experiments across diverse workloads reveal that GEMEL reduces memory usage by up to 60.7%, and improves overall accuracy by 8-39% relative to time/space sharing alone.

Rahmadi Trimananda, Weiyu Luo, Brian Demsky et al.

Event-driven architectures are broadly used for systems that must respond to events in the real world. Event-driven applications are prone to concurrency bugs that involve subtle errors in reasoning about the ordering of events. Unfortunately, there are several challenges in using existing model-checking techniques on these systems. Event-driven applications often loop indefinitely and thus pose a challenge for stateless model checking techniques. On the other hand, deploying purely stateful model checking can explore large sets of equivalent executions. In this work, we explore a new technique that combines dynamic partial order reduction with stateful model checking to support non-terminating applications. Our work is (1) the first dynamic partial order reduction algorithm for stateful model checking that is sound for non-terminating applications and (2) the first dynamic partial reduction algorithm for stateful model checking of event-driven applications. We experimented with the IoTCheck dataset: a study of interactions in smart home app pairs. This dataset consists of app pairs originated from 198 real-world smart home apps. Overall, our DPOR algorithm successfully reduced the search space for the app pairs, enabling 69 pairs of apps that did not finish without DPOR to finish and providing a 7X average speedup.

Bowen Tang, Chenggang Wu, Zhe Wang et al.

Speculative execution techniques have been a cornerstone of modern processors to improve instruction-level parallelism. However, recent studies showed that this kind of techniques could be exploited by attackers to leak secret data via transient execution attacks, such as Spectre. Many defenses are proposed to address this problem, but they all face various challenges: (1) Tracking data flow in the instruction pipeline could comprehensively address this problem, but it could cause pipeline stalls and incur high performance overhead; (2) Making side effect of speculative execution imperceptible to attackers, but it often needs additional storage components and complicated data movement operations. In this paper, we propose a label-based transparent speculation scheme called SpecBox. It dynamically partitions the cache system to isolate speculative data and non-speculative data, which can prevent transient execution from being observed by subsequent execution. Moreover, it uses thread ownership semaphores to prevent speculative data from being accessed across cores. In addition, SpecBox also enhances the auxiliary components in the cache system against transient execution attacks, such as hardware prefetcher. Our security analysis shows that SpecBox is secure and the performance evaluation shows that the performance overhead on SPEC CPU 2006 and PARSEC-3.0 benchmarks is small.

John Thorpe, Yifan Qiao, Jonathan Eyolfson et al.

A graph neural network (GNN) enables deep learning on structured graph data. There are two major GNN training obstacles: 1) it relies on high-end servers with many GPUs which are expensive to purchase and maintain, and 2) limited memory on GPUs cannot scale to today's billion-edge graphs. This paper presents Dorylus: a distributed system for training GNNs. Uniquely, Dorylus can take advantage of serverless computing to increase scalability at a low cost. The key insight guiding our design is computation separation. Computation separation makes it possible to construct a deep, bounded-asynchronous pipeline where graph and tensor parallel tasks can fully overlap, effectively hiding the network latency incurred by Lambdas. With the help of thousands of Lambda threads, Dorylus scales GNN training to billion-edge graphs. Currently, for large graphs, CPU servers offer the best performance-per-dollar over GPU servers. Just using Lambdas on top of CPU servers offers up to 2.75x more performance-per-dollar than training only with CPU servers. Concretely, Dorylus is 1.22x faster and 4.83x cheaper than GPU servers for massive sparse graphs. Dorylus is up to 3.8x faster and 10.7x cheaper compared to existing sampling-based systems.