DOCSLIB.ORG

Programmable Stream Processors

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Programmable Stream Processors COVER FEATURE Programmable Stream Processors Stream processing promises to bridge the gap between inflexible special- purpose solutions and current programmable architectures that cannot meet the computational demands of media-processing applications. Ujval J. he complexity of modern media processing, putations to occur in parallel—and require mini- Kapasi including 3D graphics, image compression, mal global communication and storage—enabling Stanford University and signal processing, requires tens to hun- data to pass directly from one ALU to the next. A dreds of billions of computations per sec- stream architecture exploits this locality and con- Scott Rixner T ond. To achieve these computation rates, currency by partitioning the communication and Rice University current media processors use special-purpose archi- storage structures to support many ALUs efficiently: tectures tailored to one specific application. Such William J. processors require significant design effort and are • operands for arithmetic operations reside in Dally thus difficult to change as media-processing appli- local register files (LRFs) near the ALUs, in cations and algorithms evolve. much the same way that special-purpose archi- Brucek The demand for flexibility in media processing tectures store and communicate data locally; Khailany motivates the use of programmable processors. • streams of data capture coarse-grained local- Jung Ho Ahn However, very large-scale integration constraints ity and are stored in a stream register file (SRF), Stanford University limit the performance of traditional programmable which can efficiently transfer data to and from architectures. In modern VLSI technology, compu- the LRFs between major computations; and Peter tation is relatively cheap—thousands of arithmetic • global data is stored off-chip only when nec- logic units that operate at multigigahertz rates can essary. Mattson 2 Reservoir Labs fit on a modestly sized 1-cm die. The problem is that delivering instructions and data to those ALUs is pro- These three explicit levels of storage form a data John D. hibitively expensive. For example, only 6.5 percent bandwidth hierarchy with the LRFs providing an Owens of the Itanium 2 die is devoted to the 12 integer and order of magnitude more bandwidth than the SRF two floating-point ALUs and their register files1; com- and the SRF providing an order of magnitude more University of California, Davis munication, control, and storage overhead consume bandwidth than off-chip storage. the remaining die area. In contrast, the more efficient This bandwidth hierarchy is well matched to the communication and control structures of a special- characteristics of modern VLSI technology, as each purpose graphics chip, such as the Nvidia GeForce4, level provides successively more storage and less enable the use of many hundreds of floating-point bandwidth. By exploiting the locality inherent in and integer ALUs to render 3D images. media-processing applications, this hierarchy stores the data at the appropriate level, enabling hundreds STREAM PROCESSING of ALUs to operate at close to their peak rate. In part, such special-purpose media processors Moreover, a stream architecture can support such are successful because media applications have a large number of ALUs in an area- and power-effi- abundant parallelism—enabling thousands of com- cient manner. Modern high-performance micro- 54 Computer Published by the IEEE Computer Society 0018-9162/03/$17.00 © 2003 IEEE Convert while ( ! Input_Image.end() ) { Luminance // input next macroblock Input_Image in = Input_Image.pop(); // generate Luminance and // Chrominance blocks outY[0..3]= gen_L_blocks(in); outC[0..1]= gen_C_blocks(in); Chrominance (a) // output new blocks Luminance.push(outY[0..3]); Chrominance.push(outC[0..1]); } struct MACROBLOCK { struct RGB_pixel { byte r,g,b; stream <MACROBLOCK> struct BLOCK { } stream <BLOCK> byte intensity[8][8]; RGB_pixel pixels[16][16]; } } stream <MACROBLOCK> Input_Image(NUM_MB); stream <BLOCK> Luminance(NUM_MB*4), Chrominance(NUM_MB*2), (b) Input_Image = Video_Feed.get_macroblocks(currpos, NUM_MB); currpos += NUM_MB; Convert(Input_Image, Luminance, Chrominance); Figure 1. Streams processors and digital signal processors continue to the Luminance and Chrominance output and a kernel from to rely on global storage and communication struc- streams, respectively. an MPEG-2 video tures to deliver data to the ALUs; these structures Streams. As Figure 1a shows, streams contain a encoder. (a) The use more area and consume more power per ALU set of elements of the same type. Stream elements Convert kernel than a stream processor. can be simple, such as a single number, or com- translates a stream plex, such as the coordinates of a triangle in 3D of macroblocks STREAMS AND KERNELS space. Streams need not be the same length—for containing RGB The central idea behind stream processing is to example, the Luminance stream has four times as pixels into streams organize an application into streams and kernels many elements as the input stream. Further, of blocks containing to expose the inherent locality and concurrency in Input_Image could contain all of the macroblocks luminance and media-processing applications. In most cases, not in an entire video frame, only a row of mac- chrominance pixels. only do streams and kernels expose desirable prop- roblocks from the frame, or even a subset of a sin- (b) A StreamC erties of media applications, but they are also a nat- gle row. In the stream code in Figure 1b, the value program expresses ural way of expressing the application. This leads of NUM_MB controls the length of the input the flow of streams to an intuitive programming model that maps stream. through kernels directly to stream architectures with tens to hun- Kernels. The Convert kernel consists of a loop textually. dreds of ALUs. that processes each element from the input stream. The body of the loop first pops an element from Example application its input stream, performs some computation on Figure 1 illustrates input and output streams and that element, and then pushes the results onto the a kernel taken from an MPEG-2 video encoder. two output streams. Figure 1a shows how a kernel operates on streams Kernels can have one or more input and output graphically, while Figure 1b shows this process in streams and perform complex calculations ranging a simplified form of StreamC, a stream program- from a few to thousands of operations per input ming language. element—one Convert implementation requires Input_Image is a stream that consists of image 6,464 operations per input macroblock to produce data from a camera. Elements of Input_Image are the six output blocks. The only external data that 16 × 16 pixel regions, or macroblocks, on which a kernel can access are its input and output streams. the Convert kernel operates. The kernel applies For example, Convert cannot directly access the the same computation to the macroblocks in data from the video feed; instead, the data must first Input_Image, decomposing each one into six 8 × 8 be organized into a stream. blocks—four luminance blocks and two 4:1 sub- Full application. A full application, such as the sampled chrominance blocks—and appends them MPEG-2 encoder, is composed of multiple streams August 2003 55 Stream Scalar Kernel Global Luminance IQDCT reference DCTQ Rate control Luminance Input_Image Variable- Video Run-level Convert length Compressed frames encoding Chrominance coding bitstream DCTQ Rate control IQDCT Chrominance reference Figure 2. MPEG-2 and kernels. This application compresses a sequence • Stream. Data are communicated between com- I-frame encoder of video images into a single bitstream consisting of putation kernels explicitly as data streams. In mapped to streams three types of frames: intracoded, predicted, and the MPEG-2 encoder, for example, the and kernels. The bidirectional. The encoder compresses I-frames Luminance and Chrominance streams use this encoder receives using only information contained in the current type of communication. a stream of macro- frame, and it compresses P- and B-frames using • Global. This level of communication is only blocks from a frame information from the current frame as well as addi- for truly global data. This is necessary for com- in the video feed as tional reference frames. municating data to and from I/O devices, as input, and the first For example, Figure 2 shows one possible map- well as for data that must persist throughout kernel (Convert) ping of the portion of the MPEG-2 encoder appli- the application. For example, the MPEG-2 I- processes this cation that encodes only I-frames into the stream- frame encoder uses this level of communica- stream. The processing model. Solid arrows represent data tion for the original input data from a video discrete cosine streams, and ovals represent computation kernels. feed and for the reference frames, which must transform (DCT) The encoder receives a stream of macroblocks persist throughout the processing of multiple kernels then from the video feed as input (Input_Image), and the video frames in the off-chip dynamic RAM operate on the first kernel (Convert) processes this input. The dis- (DRAM). output streams crete cosine transform and quantization (DCTQ) produced by kernels then operate on the output streams pro- By requiring programmers to explicitly use the Convert. duced by Convert. Upon execution of all the com- appropriate type of communication for each data Q = quantization; putation kernels, the application
Load more

Recommended publications
  • Parallel Patterns for Adaptive Data Stream Processing
    Università degli Studi di Pisa Dipartimento di Informatica Dottorato di Ricerca in Informatica Ph.D. Thesis Parallel Patterns for Adaptive Data Stream Processing Tiziano De Matteis Supervisor Supervisor Marco Danelutto Marco Vanneschi Abstract In recent years our ability to produce information has been growing steadily, driven by an ever increasing computing power, communication rates, hardware and software sensors diffusion. is data is often available in the form of continuous streams and the ability to gather and analyze it to extract insights and detect patterns is a valu- able opportunity for many businesses and scientific applications. e topic of Data Stream Processing (DaSP) is a recent and highly active research area dealing with the processing of this streaming data. e development of DaSP applications poses several challenges, from efficient algorithms for the computation to programming and runtime systems to support their execution. In this thesis two main problems will be tackled: • need for high performance: high throughput and low latency are critical re- quirements for DaSP problems. Applications necessitate taking advantage of parallel hardware and distributed systems, such as multi/manycores or cluster of multicores, in an effective way; • dynamicity: due to their long running nature (24hr/7d), DaSP applications are affected by highly variable arrival rates and changes in their workload charac- teristics. Adaptivity is a fundamental feature in this context: applications must be able to autonomously scale the used resources to accommodate dynamic requirements and workload while maintaining the desired Quality of Service (QoS) in a cost-effective manner. In the current approaches to the development of DaSP applications are still miss- ing efficient exploitation of intra-operator parallelism as well as adaptations strategies with well known properties of stability, QoS assurance and cost awareness.
    [Show full text]
  • A Middleware for Efficient Stream Processing in CUDA
    Noname manuscript No. (will be inserted by the editor) A Middleware for E±cient Stream Processing in CUDA Shinta Nakagawa ¢ Fumihiko Ino ¢ Kenichi Hagihara Received: date / Accepted: date Abstract This paper presents a middleware capable of which are expressed as kernels. On the other hand, in- out-of-order execution of kernels and data transfers for put/output data is organized as streams, namely se- e±cient stream processing in the compute uni¯ed de- quences of similar data records. Input streams are then vice architecture (CUDA). Our middleware runs on the passed through the chain of kernels in a pipelined fash- CUDA-compatible graphics processing unit (GPU). Us- ion, producing output streams. One advantage of stream ing the middleware, application developers are allowed processing is that it can exploit the parallelism inherent to easily overlap kernel computation with data trans- in the pipeline. For example, the execution of the stages fer between the main memory and the video memory. can be overlapped with each other to exploit task par- To maximize the e±ciency of this overlap, our middle- allelism. Furthermore, di®erent stream elements can be ware performs out-of-order execution of commands such simultaneously processed to exploit data parallelism. as kernel invocations and data transfers. This run-time One of the stream architecture that bene¯t from capability can be used by just replacing the original the advantages mentioned above is the graphics pro- CUDA API calls with our API calls. We have applied cessing unit (GPU), originally designed for acceleration the middleware to a practical application to understand of graphics applications.
    [Show full text]
  • AMD Accelerated Parallel Processing Opencl Programming Guide
    AMD Accelerated Parallel Processing OpenCL Programming Guide November 2013 rev2.7 © 2013 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, AMD Accelerated Parallel Processing, the AMD Accelerated Parallel Processing logo, ATI, the ATI logo, Radeon, FireStream, FirePro, Catalyst, and combinations thereof are trade- marks of Advanced Micro Devices, Inc. Microsoft, Visual Studio, Windows, and Windows Vista are registered trademarks of Microsoft Corporation in the U.S. and/or other jurisdic- tions. Other names are for informational purposes only and may be trademarks of their respective owners. OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos. The contents of this document are provided in connection with Advanced Micro Devices, Inc. (“AMD”) products. AMD makes no representations or warranties with respect to the accuracy or completeness of the contents of this publication and reserves the right to make changes to specifications and product descriptions at any time without notice. The information contained herein may be of a preliminary or advance nature and is subject to change without notice. No license, whether express, implied, arising by estoppel or other- wise, to any intellectual property rights is granted by this publication. Except as set forth in AMD’s Standard Terms and Conditions of Sale, AMD assumes no liability whatsoever, and disclaims any express or implied warranty, relating to its products including, but not limited to, the implied warranty of merchantability, fitness for a particular purpose, or infringement of any intellectual property right. AMD’s products are not designed, intended, authorized or warranted for use as compo- nents in systems intended for surgical implant into the body, or in other applications intended to support or sustain life, or in any other application in which the failure of AMD’s product could create a situation where personal injury, death, or severe property or envi- ronmental damage may occur.
    [Show full text]
  • Fine-Grained Window-Based Stream Processing on CPU-GPU Integrated
    FineStream: Fine-Grained Window-Based Stream Processing on CPU-GPU Integrated Architectures Feng Zhang and Lin Yang, Renmin University of China; Shuhao Zhang, Technische Universität Berlin and National University of Singapore; Bingsheng He, National University of Singapore; Wei Lu and Xiaoyong Du, Renmin University of China https://www.usenix.org/conference/atc20/presentation/zhang-feng This paper is included in the Proceedings of the 2020 USENIX Annual Technical Conference. July 15–17, 2020 978-1-939133-14-4 Open access to the Proceedings of the 2020 USENIX Annual Technical Conference is sponsored by USENIX. FineStream: Fine-Grained Window-Based Stream Processing on CPU-GPU Integrated Architectures Feng Zhang1, Lin Yang1, Shuhao Zhang2;3, Bingsheng He3, Wei Lu1, Xiaoyong Du1 1Key Laboratory of Data Engineering and Knowledge Engineering (MOE), and School of Information, Renmin University of China 2DIMA, Technische Universität Berlin 3School of Computing, National University of Singapore [email protected], [email protected], [email protected], [email protected], [email protected], [email protected] Abstract GPU memory via PCI-e before GPU processing, but the low Accelerating SQL queries on stream processing by utilizing bandwidth of PCI-e limits the performance of stream process- heterogeneous coprocessors, such as GPUs, has shown to be ing on GPUs. Hence, stream processing on GPUs needs to be an effective approach. Most works show that heterogeneous carefully designed to hide the PCI-e overhead. For example, coprocessors bring significant performance improvement be- prior works have explored pipelining the computation and cause of their high parallelism and computation capacity.
    [Show full text]
  • Lightsaber: Efficient Window Aggregation on Multi-Core Processors
    Research 28: Stream Processing SIGMOD ’20, June 14–19, 2020, Portland, OR, USA LightSaber: Efficient Window Aggregation on Multi-core Processors Georgios Theodorakis Alexandros Koliousis∗ Peter Pietzuch Imperial College London Graphcore Research Holger Pirk [email protected] [email protected] [email protected] [email protected] Imperial College London Abstract Invertible Aggregation (Sum) Non-Invertible Aggregation (Min) 200 Window aggregation queries are a core part of streaming ap- 140 150 Multiple TwoStacks 120 tuples/s) Multiple TwoStacks plications. To support window aggregation efficiently, stream 6 Multiple SoE 100 SlickDeque SlickDeque 80 100 SlideSide SlideSide processing engines face a trade-off between exploiting par- 60 allelism (at the instruction/multi-core levels) and incremen- 50 40 20 tal computation (across overlapping windows and queries). 0 0 Throughput (10 1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10 Existing engines implement ad-hoc aggregation and par- # Queries # Queries allelization strategies. As a result, they only achieve high Figure 1: Evaluating window aggregation queries performance for specific queries depending on the window definition and the type of aggregation function. an order of magnitude higher throughput compared to ex- We describe a general model for the design space of win- isting systems—on a 16-core server, it processes 470 million dow aggregation strategies. Based on this, we introduce records/s with 132 µs average latency. LightSaber, a new stream processing engine that balances parallelism and incremental processing when executing win- CCS Concepts dow aggregation queries on multi-core CPUs.
    [Show full text]
  • Parallel Stream Processing with MPI for Video Analytics and Data Visualization
    Parallel Stream Processing with MPI for Video Analytics and Data Visualization Adriano Vogel1 [0000−0003−3299−2641], Cassiano Rista1, Gabriel Justo1, Endrius Ewald1, Dalvan Griebler1;3[0000−0002−4690−3964], Gabriele Mencagli2, and Luiz Gustavo Fernandes1[0000−0002−7506−3685] 1 School of Technology, Pontifical Catholic University of Rio Grande do Sul, Porto Alegre - Brazil [email protected] 2 Department of Computer Science, University of Pisa, Pisa - Italy 3 Laboratory of Advanced Research on Cloud Computing (LARCC), Tr^esde Maio Faculty (SETREM), Tr^esde Maio - Brazil Abstract. The amount of data generated is increasing exponentially. However, processing data and producing fast results is a technological challenge. Parallel stream processing can be implemented for handling high frequency and big data flows. The MPI parallel programming model offers low-level and flexible mechanisms for dealing with distributed ar- chitectures such as clusters. This paper aims to use it to accelerate video analytics and data visualization applications so that insight can be ob- tained as soon as the data arrives. Experiments were conducted with a Domain-Specific Language for Geospatial Data Visualization and a Person Recognizer video application. We applied the same stream par- allelism strategy and two task distribution strategies. The dynamic task distribution achieved better performance than the static distribution in the HPC cluster. The data visualization achieved lower throughput with respect to the video analytics due to the I/O intensive operations. Also, the MPI programming model shows promising performance outcomes for stream processing applications. Keywords: parallel programming, stream parallelism, distributed pro- cessing, cluster 1 Introduction Nowadays, we are assisting to an explosion of devices producing data in the form of unbounded data streams that must be collected, stored, and processed in real-time [12].
    [Show full text]
  • Design and Implementation of an FPGA-Based Scalable Pipelined
    Design and Implementation of an FPGA-Based Scalable Pipelined Associative SIMD Processor Array with Specialized Variations for Sequence Comparison and MSIMD Operation A dissertation submitted to the Kent State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy by Hong Wang November 2006 Dissertation written by Hong Wang B.A., Lanzhou University, 1990 M.A., Kent State University, 2002 Ph.D., Kent State University, 2006 Approved by ___Robert A. Walker ___________________, Chair, Doctoral Dissertation Committee ___Johnnie W. Baker ___________________, Members, Doctoral Dissertation Committee ___Kenneth E Batcher __________________ ___Eugene C. Gartland, Jr._______________ _____________________________________ Accepted by ___Robert A. Walker___________________, Chair, Department of Computer Science ___John R. D. Stalvey__________________, Dean, College of Arts and Sciences ii Table of Contents 1 INTRODUCTION........................................................................................................... 3 1.1 Overview of Associative Computing Research......................................................... 5 1.2 Associative Computing ............................................................................................... 7 1.2.1 Associative Search and Responder Processing......................................................... 7 1.2.2 Associative Reduction for Maximum / Minimum Value ......................................... 9 1.3 Applications of a Scalable Associative Processor..................................................
    [Show full text]
  • SABER: Window-Based Hybrid Stream Processing for Heterogeneous Architectures
    SABER: Window-Based Hybrid Stream Processing for Heterogeneous Architectures Alexandros Koliousis†, Matthias Weidlich†‡, Raul Castro Fernandez†, Alexander L. Wolf†, Paolo Costa], Peter Pietzuch† †Imperial College London ‡Humboldt-Universität zu Berlin ]Microsoft Research {akolious, mweidlic, rc3011, alw, costa, prp}@imperial.ac.uk ABSTRACT offered by heterogeneous servers: besides multi-core CPUs with Modern servers have become heterogeneous, often combining multi- dozens of processing cores, such servers also contain accelerators core CPUs with many-core GPGPUs. Such heterogeneous architec- such as GPGPUs with thousands of cores. As heterogeneous servers tures have the potential to improve the performance of data-intensive have become commonplace in modern data centres and cloud offer- stream processing applications, but they are not supported by cur- ings [24], researchers have proposed a range of parallel streaming al- rent relational stream processing engines. For an engine to exploit a gorithms for heterogeneous architectures, including for joins [36,51], heterogeneous architecture, it must execute streaming SQL queries sorting [28] and sequence alignment [42]. Yet, we lack the design with sufficient data-parallelism to fully utilise all available heteroge- of a general-purpose relational stream processing engine that can neous processors, and decide how to use each in the most effective transparently take advantage of accelerators such as GPGPUs while way. It must do this while respecting the semantics of streaming executing arbitrary streaming SQL queries with windowing [7]. SQL queries, in particular with regard to window handling. The design of such a streaming engine for heterogeneous archi- tectures raises three challenges, addressed in this paper: We describe SABER, a hybrid high-performance relational stream processing engine for CPUs and GPGPUs.
    [Show full text]
  • Analyzing Efficient Stream Processing on Modern Hardware
    Analyzing Efficient Stream Processing on Modern Hardware Steffen Zeuch 2, Bonaventura Del Monte 2, Jeyhun Karimov 2, Clemens Lutz 2, Manuel Renz 2, Jonas Traub 1, Sebastian Breß 1;2, Tilmann Rabl 1;2, Volker Markl 1;2 1Technische Universitat¨ Berlin, 2German Research Center for Artificial Intelligence fi[email protected] 384 Flink ABSTRACT 420 Memory Bandwidth 288 Spark Modern Stream Processing Engines (SPEs) process large 100 71 23:99 28 Storm data volumes under tight latency constraints. Many SPEs Java HC 10 6:25 execute processing pipelines using message passing on shared- 3:6 C++ HC Streambox Throughput nothing architectures and apply a partition-based scale-out [M records/s] 1 0:26 Saber strategy to handle high-velocity input streams. Further- Optimized more, many state-of-the-art SPEs rely on a Java Virtual Ma- Figure 1: Yahoo! Streaming Benchmark (1 Node). chine to achieve platform independence and speed up system development by abstracting from the underlying hardware. a large number of computing nodes in a cluster, i.e., scale- In this paper, we show that taking the underlying hard- out the processing. These systems trade single node perfor- ware into account is essential to exploit modern hardware mance for scalability to large clusters and use a Java Virtual efficiently. To this end, we conduct an extensive experimen- Machine (JVM) as the underlying processing environment tal analysis of current SPEs and SPE design alternatives for platform independence. While JVMs provide a high level optimized for modern hardware. Our analysis highlights po- of abstraction from the underlying hardware, they cannot tential bottlenecks and reveals that state-of-the-art SPEs are easily provide efficient data access due to processing over- not capable of fully exploiting current and emerging hard- heads induced by data (de-)serialization, objects scattering ware trends, such as multi-core processors and high-speed in main memory, virtual functions, and garbage collection.
    [Show full text]
  • Streambox: Modern Stream Processing on a Multicore Machine
    StreamBox: Modern Stream Processing on a Multicore Machine Hongyu Miao and Heejin Park, Purdue ECE; Myeongjae Jeon and Gennady Pekhimenko, Microsoft Research; Kathryn S. McKinley, Google; Felix Xiaozhu Lin, Purdue ECE https://www.usenix.org/conference/atc17/technical-sessions/presentation/miao This paper is included in the Proceedings of the 2017 USENIX Annual Technical Conference (USENIX ATC ’17). July 12–14, 2017 • Santa Clara, CA, USA ISBN 978-1-931971-38-6 Open access to the Proceedings of the 2017 USENIX Annual Technical Conference is sponsored by USENIX. StreamBox: Modern Stream Processing on a Multicore Machine Hongyu Miao1, Heejin Park1, Myeongjae Jeon2, Gennady Pekhimenko2, Kathryn S. McKinley3, and Felix Xiaozhu Lin1 1Purdue ECE 2Microsoft Research 3Google Abstract Most stream processing engines are distributed be- cause they assume processing requirements outstrip the Stream analytics on real-time events has an insatiable de- capabilities of a single machine [38, 28, 32]. How- mand for throughput and latency. Its performance on a ever, modern hardware advances make a single multi- single machine is central to meeting this demand, even in core machine an attractive streaming platform. These ad- a distributed system. This paper presents a novel stream vances include (i) high throughput I/O that significantly StreamBox processing engine called that exploits the improves ingress rate, e.g., Remote Direct Memory Ac- parallelism and memory hierarchy of modern multicore cess (RDMA) and 10Gb Ethernet; (ii) terabyte DRAMs StreamBox hardware. executes a pipeline of transforms that hold massive in-memory stream processing state; over records that may arrive out-of-order. As records ar- and (iii) a large number of cores.
    [Show full text]
  • A Media Enhanced Vector Architecture for Embedded Memory Systems
    A Media-Enhanced Vector Architecture for Embedded Memory Systems Christoforos Kozyrakis Report No. UCB/CSD-99-1059 July 1999 Computer Science Division (EECS) University of California Berkeley, California 94720 A Media-Enhanced Vector Architecture for Embedded Memory Systems by Christoforos Kozyrakis Research Project Submitted to the Department of Electrical Engineering and Computer Sciences, Uni- versity of California at Berkeley, in partial satisfaction of the requirements for the degree of Master of Science, Plan II. Approval for the Report and Comprehensive Examination: Committee: Professor David A. Patterson Research Advisor (Date) ******* Professor Katherine Yelick Second Reader (Date) A Media-Enhanced Vector Architecture for Embedded Memory Systems Christoforos Kozyrakis M.S. Report Abstract Next generation portable devices will require processors with both low energy consump- tion and high performance for media functions. At the same time, modern CMOS technology creates the need for highly scalable VLSI architectures. Conventional processor architec- tures fail to meet these requirements. This paper presents the architecture of Vector IRAM (VIRAM), a processor that combines vector processing with embedded DRAM technology. Vector processing achieves high multimedia performance with simple hardware, while em- bedded DRAM provides high memory bandwidth at low energy consumption. VIRAM pro- vides ¯exible support for media data types, short vectors, and DSP features. The vector pipeline is enhanced to hide DRAM latency without using caches. The peak performance is 3.2 GFLOPS (single precision) and maximum memory bandwidth is 25.6 GBytes/s. With a target power consumption of 2 Watts for the vector pipeline and the memory system, VIRAM supports 1.6 GFLOPS/Watt. For a set of representative media kernels, VIRAM sustains on average 88% of its peak performance, outperforming conventional SIMD media extensions and DSP processors by factors of 4.5 to 17.
    [Show full text]
  • Fine-Grained Real-Time Stream Processing with Cameo
    Move Fast and Meet Deadlines: Fine-grained Real-time Stream Processing with Cameo Le Xu, University of Illinois at Urbana-Champaign; Shivaram Venkataraman, UW-Madison; Indranil Gupta, University of Illinois at Urbana-Champaign; Luo Mai, University of Edinburgh; Rahul Potharaju, Microsoft https://www.usenix.org/conference/nsdi21/presentation/xu This paper is included in the Proceedings of the 18th USENIX Symposium on Networked Systems Design and Implementation. April 12–14, 2021 978-1-939133-21-2 Open access to the Proceedings of the 18th USENIX Symposium on Networked Systems Design and Implementation is sponsored by Move Fast and Meet Deadlines: Fine-grained Real-time Stream Processing with Cameo Le Xu1∗, Shivaram Venkataraman2, Indranil Gupta1, Luo Mai3, and Rahul Potharaju4 1University of Illinois at Urbana-Champaign, 2UW-Madison, 3University of Edinburgh, 4Microsoft Abstract Resource provisioning in multi-tenant stream processing sys- tems faces the dual challenges of keeping resource utilization high (without over-provisioning), and ensuring performance isolation. In our common production use cases, where stream- ing workloads have to meet latency targets and avoid breach- ing service-level agreements, existing solutions are incapable Figure 1: Slot-based system (Flink), Simple Actor system (Or- of handling the wide variability of user needs. Our framework leans), and our framework Cameo. called Cameo uses fine-grained stream processing (inspired by actor computation models), and is able to provide high latency. Others wish to maximize throughput under limited re- resource utilization while meeting latency targets. Cameo dy- sources, and yet others desire high resource utilization. Violat- namically calculates and propagates priorities of events based ing such user expectations is expensive, resulting in breaches on user latency targets and query semantics.
    [Show full text]