Marc Snir

The increasing complexity of HPC architectures and the growing adoption of irregular scientific algorithms demand efficient support for asynchronous, multithreaded communication. This need is especially pronounced with Asynchronous Many-Task (AMT) systems. This communication pattern was not a consideration during the design of the original MPI specification. The MPI community has recently introduced several extensions to address these evolving requirements. This work evaluates two such extensions, the Virtual Communication Interface (VCI) and the Continuation extensions, in the context of an established AMT runtime HPX. We begin by using an MPI-level microbenchmark, modeled from HPX's low-level communication mechanism, to measure the peak performance potential of these extensions. We then integrate them into HPX to evaluate their effectiveness in real-world scenarios. Our results show that while these extensions can enhance performance compared to standard MPI, areas for improvement remain. The current continuation proposal limits the maximum multithreaded message rate achievable in the multi-VCI setting. Furthermore, the recommended one-VCI-per-thread mode proves ineffective in real-world systems due to the attentiveness problem. These findings underscore the importance of improving intra-VCI threading efficiency to achieve scalable multithreaded communication and fully realize the benefits of recent MPI extensions.

High-performance computing (HPC) applications generate and consume substantial amounts of data, typically managed by parallel file systems. These applications access file systems either through the POSIX interface or by using highlevel I/O libraries. While the POSIX consistency model remains dominant in HPC, emerging file systems and popular I/O libraries increasingly adopt alternative consistency models that relax semantics in various ways, creating significant challenges for correctness and portability. This paper addresses these challenges by proposing a trace-driven I/O consistency verification workflow, implemented in our open-source tool, VerifyIO, which collects execution traces, detects data conflicts, and verifies proper synchronization against specified consistency models. Our extensive evaluation of 91 test case executions across three widely used I/O libraries with four I/O consistency models reveals critical consistency issues at both application and implementation levels.

Cloud platforms have been virtualizing storage devices like flash-based solid-state drives (SSDs) to make effective use of storage resources. They enable either software-isolated instance or hardware-isolated instance for facilitating the storage sharing between multi-tenant applications. However, for decades, they have to combat the fundamental tussle between the performance isolation and resource utilization. They suffer from either long tail latency caused by weak isolation or low storage utilization caused by strong isolation. In this paper, we present FleetIO, a learning-based storage virtualization framework that employs reinforcement learning (RL) for managing virtualized SSDs. FleetIO explores the unique features of RL to handle the dynamic changes of application workloads and storage states, and integrates the storage scheduling into the RL decision-making process. It achieves both performance isolation and improved storage utilization by enabling dynamic fine-grained storage harvesting across collocated application instances, while minimizing its negative impact on their service-level objectives (SLOs). FleetIO clusters workloads into different types (e.g., latency-sensitive and bandwidth-intensive) based on the collected I/O traces at runtime, and fine-tunes the RL reward functions for each type of workloads. We implement FleetIO on a real programmable SSD board and evaluate it with diverse cloud applications. We show that FleetIO improves the overall storage utilization of the shared SSD by up to 1.4×, and decreases the tail latency of I/O requests by 1.5× on average, compared to the state-of-the-art storage sharing approaches.

Asynchronous Many-Task Systems (AMTs) exhibit different communication patterns from traditional High-Performance Computing (HPC) applications, characterized by asynchrony, concurrency, and multithreading. Existing communication libraries usually do not support AMTs' communication requirements in the most direct and efficient ways. The Lightweight Communication Interface (LCI) is an experimental communication library aiming to push for efficient communication support for AMTs. This paper presents the design for a new LCI C++ interface and its rationale. With a new C++ \emph{objectized flexible functions} idiom, the new interface aims for the following features: (a) a concise but expressive interface for all common point-to-point communication primitives and completion mechanisms, (b) a fine-grained resource mapping scheme for library interoperation, multithreaded performance isolation, and flexibility (c) a set of optional parameters and overridable classes for users to incrementally fine-tune the runtime behavior.

Asynchronous Many-Task (AMT) systems offer a potential solution for efficiently programming complicated scientific applications on extreme-scale heterogeneous architectures. However, they exhibit different communication needs from traditional bulk-synchronous parallel (BSP) applications, posing new challenges for underlying communication libraries. This work systematically studies the communication needs of AMTs and explores how communication libraries can be structured to better satisfy them through a case study of a real-world AMT system, HPX. We first examine its communication stack layout and formalize the communication abstraction that underlying communication libraries need to support. We then analyze its current MPI backend (parcelport) and identify four categories of needs that are not typical in the BSP model and are not well covered by the MPI standard. To bridge these gaps, we design from the native network layer and incorporate various techniques, including one-sided communication, queue-based completion notification, explicit progressing, and different ways of resource contention mitigation, in a new parcelport with an experimental communication library, LCI. Overall, the resulting LCI parcelport outperforms the existing MPI parcelport with up to 50x in microbenchmarks and 2x in a real-world application. Using it as a testbed, we design LCI parcelport variants to quantify the performance contributions of each technique. This work combines conceptual analysis and experiment results to offer a practical guideline for the future development of communication libraries and AMT communication layers.

This paper presents Recorder, a parallel I/O tracing tool designed to capture comprehensive I/O information on HPC applications. Recorder traces I/O calls across various I/O layers, storing all function parameters for each captured call. The volume of stored information scales linearly the application's execution scale. To address this, we present a sophisticated pattern-recognition-based compression algorithm. This algorithm identifies and compresses recurring I/O patterns both within individual processes and across multiple processes, significantly reducing space and time overheads. We evaluate the proposed compression algorithm using I/O benchmarks and real-world applications, demonstrating that Recorder can store more information while requiring approximately 12x less storage space compared to its predecessor. Notably, for applications with typical parallel I/O patterns, Recorder achieves a constant trace size regardless of execution scale. Additionally, a comparison with the profiling tool Darshan shows that Recorder captures detailed I/O information without incurring substantial overhead. The richer data collected by Recorder enables new insights and facilitates more in-depth I/O studies, offering valuable contributions to the I/O research community.

Modular data centers (MDCs) that can be placed right at the energy farms and powered mostly by renewable energy, is a flexible and effective approach to lowering the carbon footprint of data centers. However, the main challenge of using renewable energy is the high variability of power produced, which implies large volatility in powering computing resources at MDCs, and degraded application performance due to the task evictions and migrations. This causes challenges for platform operators to decide the MDC deployment. To this end, we present SkyBox, a framework that employs a learning-based approach for platform operators to explore the efficient use of renewable energy with MDC deployment across geographical regions. SkyBox is driven by the insights based on our study of real-world power traces from a variety of renewable energy farms - the predictable production of renewable energy and the complementary nature of energy production patterns across different renewable energy sources and locations. With these insights, SkyBox uses the coefficient of variation metric to select the qualified renewable farms, it can identify a set of farms with complementary energy production patterns with a subgraph identification algorithm. After that, SkyBox enables smart workload placement and migrations to further tolerate the power variability. Our experiments with real power traces and datacenter workloads show that SkyBox has the lowest carbon emissions compared with existing approaches. SkyBox also minimizes the negative impact of the power variability on cloud applications, enabling it an effective solution of utilizing renewable energy for modern data centers.