FAME: FAirly MEasuring Multithreaded Architectures

GPUs are being increasingly adopted as compute accelerators in many domains, spanning environments from mobile systems to cloud computing. These systems are usually running multiple applications, from one or several users. However GPUs do not provide the support for resource sharing traditionally expected in these scenarios. Thus, such systems are unable to provide key multiprogrammed workload requirements, such as responsiveness, fairness or quality of service.In this paper, we propose a set of hardware extensions that allow GPUs to efficiently support multiprogrammed GPU workloads. We argue for preemptive multitasking and design two preemption mechanisms that can be used to implement GPU scheduling policies. We extend the architecture to allow concurrent execution of GPU kernels from different user processes and implement a scheduling policy that dynamically distributes the GPU cores among concurrently running kernels, according to their priorities. We extend the NVIDIA GK110 (Kepler) like GPU architecture with our proposals and evaluate them on a set of multiprogrammed workloads with up to eight concurrent processes. Our proposals improve execution time of high-priority processes by 15.6x, the average application turnaround time between 1.5x to 2x, and system fairness up to 3.4x.

show abstract

“…Statistics are gathered only for the completed executions and then averaged. This methodology is based on [36] and [37].…”

Section: Methodsmentioning

confidence: 99%

Enabling preemptive multiprogramming on GPUs

Tanasic

Gelado

Cabezas

et al. 2014

SIGARCH Comput. Archit. News

Self Cite

View full text Add to dashboard Cite

show abstract

“…Multi-program Benchmark Definition FAME [21] constructs multi-program benchmarks by repeating the executions of the single programs enough times to ensure that a steady-state behavior is achieved. In our benchmark definition terminology, FAME proposes setting T to be "steady state behavior is obtained."…”

Section: Impact Of S 0 Orderingmentioning

confidence: 99%

“…Approaches to variable instruction count sampling include "run both programs, possibly starting from checkpoints, until the sum of their committed instruction counts reaches a specified target" (e.g., [16]), "run until both programs reach a minimum number of instructions or until one program reaches a maximum number of completed instructions" (e.g. [11]), and "run until each program has reached a minimum number of instructions, restarting the faster program as needed" (i.e., FAME [21]). The samples from these methodologies represent benchmarks similar to those in the TM category; the ratios of their constituent programs vary depending on the underlying micro-architecture.…”

Section: B Sampling Multi-program Benchmarksmentioning

confidence: 99%

Multi-program benchmark definition

Jacobvitz

Hilton

Sorin

2015

2015 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)

View full text Add to dashboard Cite

Abstract-Although definition of single-program benchmarks is relatively straight-forward-a benchmark is a program plus a specific input-definition of multi-program benchmarks is more complex. Each program may have a different runtime and they may have different interactions depending on how they align with each other. While prior work has focused on sampling multiprogram benchmarks, little attention has been paid to defining the benchmarks in their entirety.In this work, we propose a four-tuple that formally defines multi-program benchmarks in a well-defined way. We then examine how four different classes of benchmarks created by varying the elements of this tuple align with real-world use-cases. We evaluate the impact of these variations on real hardware, and see drastic variations in results between different benchmarks constructed from the same programs. Notable differences include significant speedups versus slowdowns (e.g., +57% vs -5% or +26% vs -18%), and large differences in magnitude even when the results are in the same direction (e.g., 67% versus 11%).

show abstract

“…This paper uses the FAME (FAirly MEasuring Multithreaded Architectures) methodology [24] [25]. In [24] the authors state that the average accumulated IPC of a program is representative if it is similar to the IPC of that program when the workload reaches a steady state.…”

Section: Running the Experimentsmentioning

confidence: 99%