scalable data processing at network transfer rates with
play

Scalable Data Processing at Network transfer rates with nCorium - PowerPoint PPT Presentation

Oct 28, 2022 •45 likes •125 views

Scalable Data Processing at Network transfer rates with nCorium Compute in Memory Modules Suresh Devalapalli, Brett Neuman, Arvindh Lalam 11/18/2019 1 OSS Layer Bottlenecks in HPC Facilities 10,000 000s s server ers 10s of Peta

  1. Scalable Data Processing at Network transfer rates with nCorium Compute in Memory Modules Suresh Devalapalli, Brett Neuman, Arvindh Lalam 11/18/2019 1

  2. OSS Layer Bottlenecks in HPC Facilities 10,000’ 000’s s server ers • 10s of Peta Bytes of data moved in and out of Storage layer Compu pute L e Layer er • Challenges: OSS SS 1. Data processing on CPU unable to keep up with increase in network and storage speeds 2. I/O throughput per node is limited Storage L e Layer er 11/18/2019 2

  3. nCorium Compute in Memory Module • DIMM form-factor • Reconfigurable compute cores on the DIMM • Off-load functions via libraries: • Compression • Erasure • Encryption • Video Transcoding • ….etc nCIMM Array in the system • Data Stream based processing 11/18/2019 3

  4. nCIMMs + nC-Grid = Scalable nCorium-OSS • nCIMM to off-load data processing • nC-Grid to open wider I/O 1 st Gen Server With 5 nCSTs • Scalable Architecture: nCST SSD SSD SSD SSD  Number of nCIMMs EDR (Storage Target) SSD SSD SSD SSD  Functions to support nCST SSD SSD SSD SSD CPU EDR (Storage Target) SSD SSD SSD SSD nCIMM nCIMM nCST SSD SSD SSD SSD EDR nCIMM (Storage Target) SSD SSD SSD SSD nCIMM nCIMM nCST SSD SSD SSD SSD EDR nCIMM (Storage Target) SSD SSD SSD SSD nCIMM nCIMM nCST SSD SSD SSD SSD nCIMM EDR (Storage Target) SSD SSD SSD SSD nCIMM Management Network Data Network 11/18/2019 4

  5. Performance of Gen-1 nCorium OSS: Q’1 19 RDMA 5x for RDMA + Erasure 60 3.4X RDMA RDMA + Encode 50.0 50 3.4x for RDMA+ Erasure RDMA + SSD 41.1 RDMA + Encode + SSD 40 Througbput (GBps) 34.3 30.8 28.6 30 26.3 22.5 22.2 21.1 20.7 Performance compared with Intel Gold CPU based 18.4 20 system 15.9 15.0 14.6 11 10.8 10.5 9.8 9 10 7.5 7.5 7.2 5.5 4.9 Gen-1 released in Q1’19 0 Ref. Platform 1-nc STs 2-nc STs 3-nc STs 4-nc STs 5-nc STs Number of nC-STs 11/18/2019 5

  6. Performance of Gen-2 nCorium OSS • Improved reconfigurable cores ~10x faster • More off-load functionality • Higher Power efficiency Ref. Platform vs nCorium G1 and nCorium G2 12.0 9.9 10.0 Normalized to Ref. Platform 8.0 7.5 7.2 Gen-2 now available 7.0 6.0 4.7 4.5 3.8 3.7 4.0 RDMA RDMA+ENC RDMA+ENC+SSD RDMA+COMP+ENC+SSD Reference 11 9 7.2 5 nCorium G1 50 34.3 26.3 23.5 2.0 1.0 1.0 1.0 1.0 nCorium G2 76.5 67.5 52 49.4 Numbers extrapolated from initial data 0.0 RDMA RDMA+ENC RDMA+ENC+SSD RDMA+COMP+ENC+SSD Reference nCorium G1 nCorium G2 11/18/2019 6

  7. Ongoing Work • Parallel file system integration • Other Applications • Lustre Key-Value Storage  • BeeGFS Video Transcoding  Inferencing  Graph Analytics   Cyber Security Content Distribution  11/18/2019 7

  8. Demos / Exhibits StarLight: Booth 993 Thank you! EchoStreams: Booth 1490 11/18/2019 8

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

What is Scalable Data Processing? S CALABLE DATA P ROCES S IN G IN R Michael J. Kane and Simon

What is Scalable Data Processing? S CALABLE DATA P ROCES S IN G IN R Michael J. Kane and Simon

What is Scalable Data Processing? S CALABLE DATA P ROCES S IN G IN R Michael J. Kane and Simon Urbanek Instructors, DataCamp In this course .. Work with data that is too large for your computer Write Scalable code Import and process data in

593 views • 34 slides
Enabling Scalable Parallel Processing of Venus/OMNeT++ Network Models on the IBM Blue Gene/Q

Enabling Scalable Parallel Processing of Venus/OMNeT++ Network Models on the IBM Blue Gene/Q

Enabling Scalable Parallel Processing of Venus/OMNeT++ Network Models on the IBM Blue Gene/Q Supercomputer Chris Carothers, Philip Heidelberger, Elsa Gonsiorowski and German Herrera, Justin LaPre Cyriel Minkenberg and Center for

519 views • 26 slides
DSC 102 Systems for Scalable Analytics Arun Kumar Topic 3: Parallel and Scalable Data

DSC 102 Systems for Scalable Analytics Arun Kumar Topic 3: Parallel and Scalable Data

DSC 102 Systems for Scalable Analytics Arun Kumar Topic 3: Parallel and Scalable Data Processing Ch. 9.4, 12.2, 14.1.1, 14.6, 22.1-22.3, 22.4.1, 22.8 of Cow Book Ch. 5, 6.1, 6.3, 6.4 of MLSys Book 1 Q: Why bother with large-scale data?

1.66k views • 124 slides
A Scalable Processing-in-Memory Accelerator for Parallel Graph Processing Seminar on Computer

A Scalable Processing-in-Memory Accelerator for Parallel Graph Processing Seminar on Computer

A Scalable Processing-in-Memory Accelerator for Parallel Graph Processing Seminar on Computer Architecture Roknoddin Azizibarzoki Junwhan Ahn, Sungpack Hong* Sungjoo Yoo, Onur Mutlu+ Kiyoung Choi Seoul National University *Oracle Labs

646 views • 41 slides
Data Management Network transfers Network data transfers Not everyone needs to transfer large

Data Management Network transfers Network data transfers Not everyone needs to transfer large

Data Management Network transfers Network data transfers Not everyone needs to transfer large amounts of data on and off a HPC service Sometimes data is created and consumed on the same service. If you do need to move large amounts of

336 views • 32 slides
Jumpgate: In-Network Processing as a Service for Data Analytics Craig Mustard, Fabian Ruffy, Anny

Jumpgate: In-Network Processing as a Service for Data Analytics Craig Mustard, Fabian Ruffy, Anny

Jumpgate: In-Network Processing as a Service for Data Analytics Craig Mustard, Fabian Ruffy, Anny Gakhokidze, Ivan Beschastnikh, Alexandra Fedorova University of British Columbia 1 In-Network Processing Can Accelerate Data Analytics Switches

325 views • 15 slides
CMS Data Transfer tests towards LHC data taking CMS Data Transfer tests towards LHC data taking D

CMS Data Transfer tests towards LHC data taking CMS Data Transfer tests towards LHC data taking D

CMS Data Transfer tests towards LHC data taking CMS Data Transfer tests towards LHC data taking D Bonacorsi CMS Facilities Infrastructure Operations INFN CNAF Bologna Italy On behalf of the CMS experiment

766 views • 48 slides
When simplicity becomes complex On the road to a scalable and dynamic SURFnet7 network Terena

When simplicity becomes complex On the road to a scalable and dynamic SURFnet7 network Terena

When simplicity becomes complex On the road to a scalable and dynamic SURFnet7 network Terena Network Architects Workshop, Copenhagen Wouter Huisman What do we want from a network? Scalable Flexible Cost efficient Endusers

347 views • 31 slides
A Scalable Algorithm for Radiative Heat Transfer Using Reverse Monte Carlo Ray Tracing Alan

A Scalable Algorithm for Radiative Heat Transfer Using Reverse Monte Carlo Ray Tracing Alan

A Scalable Algorithm for Radiative Heat Transfer Using Reverse Monte Carlo Ray Tracing Alan Humphrey 1 , Todd Harman 1 , Martin Berzins 1 , and Phillip Smith 1 University of Utah, Salt Lake City, USA Abstract. Radiative heat transfer is an

392 views • 18 slides
STORM AND LOW-LATENCY PROCESSING www.inf.ed.ac.uk Low latency processing Similar to data

STORM AND LOW-LATENCY PROCESSING www.inf.ed.ac.uk Low latency processing Similar to data

STORM AND LOW-LATENCY PROCESSING www.inf.ed.ac.uk Low latency processing Similar to data stream processing, but with a twist Data is streaming into the system (from a database, or a network stream, or an HDFS file, or ) We want to

573 views • 31 slides
Scalable Methods for the Analysis of Network-Based Data Principal Investigator: Professor

Scalable Methods for the Analysis of Network-Based Data Principal Investigator: Professor

Scalable Methods for the Analysis of Network-Based Data Principal Investigator: Professor Padhraic Smyth Department of Computer Science University of California Irvine Slides online at www.datalab.uci.edu/muri P. Smyth: Networks MURI Project

810 views • 53 slides
The Nomadic Network Providing Secure, Scalable and Manageable Roaming, Remote and Wireless Data

The Nomadic Network Providing Secure, Scalable and Manageable Roaming, Remote and Wireless Data

The Nomadic Network Providing Secure, Scalable and Manageable Roaming, Remote and Wireless Data Services Josh Howlett & Nick Skelton Information Services, University of Bristol TNC 2003 Background 1999-2000: new technologies

1.22k views • 52 slides
Scalable Interconnection Networks 1 Scalable, High Performance Network At Core of Parallel

Scalable Interconnection Networks 1 Scalable, High Performance Network At Core of Parallel

Scalable Interconnection Networks 1 Scalable, High Performance Network At Core of Parallel Computer Architecture Requirements and trade-offs at many levels Elegant mathematical structure Deep relationships to algorithm structure

691 views • 52 slides
Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California,

Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California,

Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California, Irvine Project Meeting August 25 th 2009 Principal Investigator: Padhraic Smyth Goals for Todays Meeting Introductions and brief review of

998 views • 32 slides
Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California,

Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California,

Scalable Methods for the Analysis of Network-Based Data MURI Project: University of California, Irvine Annual Review Meeting December 8 th 2009 Principal Investigator: Padhraic Smyth Todays Meeting Goals Review our research

553 views • 32 slides
Using sampled social network data to estimate adult death rates Dennis M. Feehan UC Berkeley

Using sampled social network data to estimate adult death rates Dennis M. Feehan UC Berkeley

Using sampled social network data to estimate adult death rates Dennis M. Feehan UC Berkeley Joint with: Matthew J. Salganik (Princeton), Mary Mahy (UNAIDS), Aline Umubyeyi (U. of Rwanda), Wolfgang Hladik (CDC) Source: Mikkelsen et al (2015),

1.14k views • 70 slides
Analytical Query Processing Marco Serafini COMPSCI 532 Lecture 7 Announcement Midterm date

Analytical Query Processing Marco Serafini COMPSCI 532 Lecture 7 Announcement Midterm date

Analytical Query Processing Marco Serafini COMPSCI 532 Lecture 7 Announcement Midterm date and location confirmed October 22 at 7-9pm in ILC S331 2 2 MapReduce vs. DBMSs 3 Advantages of DBMSs Abstract data representation

683 views • 26 slides
Challenges for Data Driven Systems Eiko Yoneki University of Cambridge Computer Laboratory Data

Challenges for Data Driven Systems Eiko Yoneki University of Cambridge Computer Laboratory Data

Challenges for Data Driven Systems Eiko Yoneki University of Cambridge Computer Laboratory Data Centric Systems and Networking Emergence of Big Data Shift of Communication Paradigm From end-to-end to data centric Data as

767 views • 17 slides
Block-Quantized Kernel Matrix for Fast Spectral Embedding Kai Zhang James T. Kwok Department of

Block-Quantized Kernel Matrix for Fast Spectral Embedding Kai Zhang James T. Kwok Department of

Block-Quantized Kernel Matrix for Fast Spectral Embedding Kai Zhang James T. Kwok Department of Computer Science and Engineering The Hong Kong University of Science and Technology Hong Kong Introduction The Proposed Method Experiments

429 views • 31 slides
T-61.3050 Machine Learning: Basic Principles Clustering Kai Puolam aki Laboratory of Computer

T-61.3050 Machine Learning: Basic Principles Clustering Kai Puolam aki Laboratory of Computer

Dimensionality Reduction Clustering T-61.3050 Machine Learning: Basic Principles Clustering Kai Puolam aki Laboratory of Computer and Information Science (CIS) Department of Computer Science and Engineering Helsinki University of

826 views • 43 slides
AIMS CDT - Signal Processing Michaelmas Term 2020 Xiaowen Dong Department of Engineering Science

AIMS CDT - Signal Processing Michaelmas Term 2020 Xiaowen Dong Department of Engineering Science

AIMS CDT - Signal Processing Michaelmas Term 2020 Xiaowen Dong Department of Engineering Science Outline Lectures - Monday-Thursday 10:00-12:00 - slides: http://www.robots.ox.ac.uk/~xdong/teaching.html Lab sessions - Tuesday &

866 views • 76 slides
Processing Data from Files n So far: n Inputs : n from user n

Processing Data from Files n So far: n Inputs : n from user n

Processing Data from Files n So far: n Inputs : n from user n "hard-wired" into program n Outputs : n "printing" on the screen n In practice, usually: n Input from file n Output to file 1

511 views • 15 slides
Imagine: Media Processing with Streams Brucek Khailany et al. and a little bit of Evaluating

Imagine: Media Processing with Streams Brucek Khailany et al. and a little bit of Evaluating

Imagine: Media Processing with Streams Brucek Khailany et al. and a little bit of Evaluating the Imagine Stream Architecture Jung Ho Ahn et al. Presented by Dan Amelang Background Digital media processing has become pervasive

177 views • 14 slides
Making Middleboxes Someone Elses Problem: Network Processing as a Cloud Service Justine

Making Middleboxes Someone Elses Problem: Network Processing as a Cloud Service Justine

Making Middleboxes Someone Elses Problem: Network Processing as a Cloud Service Justine Sherry*, Shaddi Hasan*, Colin Scott*, Arvind Krishnamurthy , Sylvia Ratnasamy*, and Vyas Sekar * Typical Enterprise Networks Internet

904 views • 52 slides

More recommend