An official website of the United States government
Concurrent kernel execution on GPU has proven an effective technique to improve system throughput by maximizing the resource utilization. In order to increase programmability and meet the increasing memory requirements of data-intensive applications, current GPUs support Unified Virtual Memory (UVM), which provides a virtual memory abstraction with demand paging. By allowing applications to oversubscribe GPU memory, UVM provides increased opportunities to share GPU resources across applications. However, in the presence of applications with competing memory requirements, GPU sharing can lead to performance degradation due to thrashing. NVIDIA's Multiple Process Service (MPS) offers the capability to space share bare metal GPUs, thereby enabling cluster workload managers, such as Slurm, to share a single GPU across MPI ranks with limited control over resource partitioning. However, it is not possible to preempt, schedule, or throttle a running GPU process through MPS. These features would enable new OS-managed scheduling policies to be implemented for GPU kernels to dynamically handle resource contention and offer consistent performance. The contribution of this paper is two-fold. We first show how memory oversubscription can impact the performance of concurrent GPU applications. Then, we propose three methods to transparently mitigate memory interference through kernel preemption and scheduling policies. To implement our policies, we develop our own runtime system (PILOT) to serve as an alternative to NVIDIA's MPS. In the presence of memory over-subscription, we noticed a dramatic improvement in the overall throughput when using our scheduling policies and runtime hints.
more »
« less
MASK: Redesigning the GPU Memory Hierarchy to Support Multi-Application Concurrency
Graphics Processing Units (GPUs) exploit large amounts of thread-level parallelism to provide high instruction throughput and to efficiently hide long-latency stalls. The resulting high throughput, along with continued programmability improvements, have made GPUs an essential computational resource in many domains. Applications from different domains can have vastly different compute and memory demands on the GPU. In a large-scale computing environment, to efficiently accommodate such wide-ranging demands without leaving GPU resources underutilized, multiple applications can share a single GPU, akin to how multiple applications execute concurrently on a CPU. Multi-application concurrency requires several support mechanisms in both hardware and software. One such key mechanism is virtual memory, which manages and protects the address space of each application. However, modern GPUs lack the extensive support for multi-application concurrency available in CPUs, and as a result suffer from high performance overheads when shared by multiple applications, as we demonstrate. We perform a detailed analysis of which multi-application concurrency support limitations hurt GPU performance the most. We find that the poor performance is largely a result of the virtual memory mechanisms employed in modern GPUs. In particular, poor address translation performance is a key obstacle to efficient GPU sharing. State-of-the-art address translation mechanisms, which were designed for single-application execution, experience significant inter-application interference when multiple applications spatially share the GPU. This contention leads to frequent misses in the shared translation lookaside buffer (TLB), where a single miss can induce long-latency stalls for hundreds of threads. As a result, the GPU often cannot schedule enough threads to successfully hide the stalls, which diminishes system throughput and becomes a first-order performance concern. Based on our analysis, we propose MASK, a new GPU framework that provides low-overhead virtual memory support for the concurrent execution of multiple applications. MASK consists of three novel address-translation-aware cache and memory management mechanisms that work together to largely reduce the overhead of address translation: (1) a token-based technique to reduce TLB contention, (2) a bypassing mechanism to improve the effectiveness of cached address translations, and (3) an application-aware memory scheduling scheme to reduce the interference between address translation and data requests. Our evaluations show that MASK restores much of the throughput lost to TLB contention. Relative to a state-of-the-art GPU TLB, MASK improves system throughput by 57.8%, improves IPC throughput by 43.4%, and reduces application-level unfairness by 22.4%. MASK's system throughput is within 23.2% of an ideal GPU system with no address translation overhead.
more »
« less
- Award ID(s):
- 1750667
- PAR ID:
- 10082129
- Date Published:
- Journal Name:
- Proceedings of the Twenty-Third International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS '18)
- Page Range / eLocation ID:
- 503 to 518
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Graphics processing units (GPUs) are becoming default accelerators in many domains such as high-performance computing (HPC), deep learning, and virtual/augmented reality. Recently, GPUs have also shown significant speedups for a variety of security-sensitive applications such as encryptions. These speedups have largely benefited from the high memory bandwidth and compute throughput of GPUs. One of the key features to optimize the memory bandwidth consumption in GPUs is intra-warp memory access coalescing, which merges memory requests originating from different threads of a single warp into as few cache lines as possible. However, this coalescing feature is also shown to make the GPUs prone to the correlation timing attacks as it exposes the relationship between the execution time and the number of coalesced accesses. Consequently, an attacker is able to correctly reveal an AES private key via repeatedly gathering encrypted data and execution time on a GPU. In this work, we propose a series of defense mechanisms to alleviate such timing attacks by carefully trading off performance for improved security. Specifically, we propose to randomize the coalescing logic such that the attacker finds it hard to guess the correct number of coalesced accesses generated. To this end, we propose to randomize: a) the granularity (called as subwarp) at which warp threads are grouped together for coalescing, and b) the threads selected by each subwarp for coalescing. Such randomization techniques result in three mechanisms: fixed-sized subwarp (FSS), random-sized subwarp (RSS), and random-threaded subwarp (RTS). We find that the combination of these security mechanisms offers 24- to 961-times improvement in the security against the correlation timing attacks with 5 to 28% performance degradation. Online copy: http://adwaitjog.github.io/docs/pdf/rcoal-hpca18.pdfmore » « less
-
As the volume of data processed by applications has increased, considerable attention has been paid to data address translation overheads, leading to the widespread use of larger page sizes (“superpages”) and multi-level translation lookaside buffers (TLBs). However, far less attention has been paid to instruction address translation and its relation to TLB and pipeline structure. In prior work, we quantified the impact of using code superpages on a variety of widely used applications, ranging from compilers to web user-interface frameworks, and the impact of sharing page table pages for executables and shared libraries. Within this article, we augment those results by first uncovering the effects that microarchitectural differences between Intel Skylake and AMD Zen+, particularly their different TLB organizations, have on instruction address translation overhead. This analysis provides some key insights into the microarchitectural design decisions that impact the cost of instruction address translation. First, a lower-level (level 2) TLB that has both instruction and data mappings competing for space within the same structure allows better overall performance and utilization when using code superpages. Code superpages not only reduce instruction address translation overhead but also indirectly reduce data address translation overhead. In fact, for a few applications, the use of just a few code superpages has a larger impact on overall performance than the use of a much larger number of data superpages. Second, a level 1 (L1) TLB with separate structures for different page sizes may require careful tuning of the superpage promotion policy for code, and a correspondingly suboptimal utilization of the level 2 TLB. In particular, increasing the number of superpages when the size of the L1 superpage structure is small may result in more L1 TLB misses for some applications. Moreover, on some microarchitectures, the cost of these misses can be highly variable, because replacement is delayed until all of the in-flight instructions mapped by the victim entry are retired. Hence, more superpage promotions can result in a performance regression. Finally, our findings also make a case for first-class OS support for superpages on ordinary files containing executables and shared libraries, as well as a more aggressive superpage policy for code.more » « less
-
An increasing number of applications benefit from heterogeneous hardware accelerators. Such accelerators often require the application to manually manage memory buffers on devices and transfer data between host and device buffers. A programming model that unifies the virtual address space across the host and devices is appealing because it enables automatic memory transfers and simplifies application-level programming. However, the automatic memory transfers can sometimes be redundant, which decreases performance. NVIDIA’s UVM (unified virtual memory) driver provides a unified virtual address space for CPU-GPU programming. This paper identifies redundant memory transfers (RMTs) as a common performance issue with UVM. To address this issue, this paper proposes a data discard directive, and evaluates two implementations of that directive, UvmDiscard and UvmDiscardLazy. This directive exploits application-level knowledge to avoid RMTs. The implementations were integrated with NVIDIA’s open-source UVM driver to demonstrate their usefulness on real-world CUDA UVM applications. For example, the use of the discard directive increases training throughput by 61.2% on a large deep learning application that oversubscribes GPU memory.more » « less
-
null (Ed.)Error-bounded lossy compression is a state-of-the-art data reduction technique for HPC applications because it not only significantly reduces storage overhead but also can retain high fidelity for postanalysis. Because supercomputers and HPC applications are becoming heterogeneous using accelerator-based architectures, in particular GPUs, several development teams have recently released GPU versions of their lossy compressors. However, existing state-of-the-art GPU-based lossy compressors suffer from either low compression and decompression throughput or low compression quality. In this paper, we present an optimized GPU version, cuSZ, for one of the best error-bounded lossy compressors-SZ. To the best of our knowledge, cuSZ is the first error-bounded lossy compressor on GPUs for scientific data. Our contributions are fourfold. (1) We propose a dual-quantization scheme to entirely remove the data dependency in the prediction step of SZ such that this step can be performed very efficiently on GPUs. (2) We develop an efficient customized Huffman coding for the SZ compressor on GPUs. (3) We implement cuSZ using CUDA and optimize its performance by improving the utilization of GPU memory bandwidth. (4) We evaluate our cuSZ on five real-world HPC application datasets from the Scientific Data Reduction Benchmarks and compare it with other state-of-the-art methods on both CPUs and GPUs. Experiments show that our cuSZ improves SZ's compression throughput by up to 370.1x and 13.1x, respectively, over the production version running on single and multiple CPU cores, respectively, while getting the same quality ofmore » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3173162.3173169
