Burst Buffer Simulation In Dragonfly Network Jian Peng, Michael - PowerPoint PPT Presentation

Aug 17, 2022 •113 likes •280 views

Burst Buffer Simulation In Dragonfly Network Jian Peng, Michael Lang Illinois Institute of Technology, Los Alamos National Laboratory Purpose: Residing in the compute node network, and using solid state drives (SSD), burst buffers bring a

Burst Buffer Simulation In Dragonfly Network Jian Peng, Michael Lang Illinois Institute of Technology, Los Alamos National Laboratory
Purpose: • Residing in the compute node network, and using solid state drives (SSD), burst buffers bring a significant I/O performance boost compared with traditional external hard disk drive (HDD) storage system. • bottleneck of fully leveraging burst buffers still remains unknown. • Sharing network • Limited burst buffer number • Due to large system scales, it is usually too expensive to change system setup and configurations.
Trinity System Overview : • System Level • Cabinet Level • Chassis Level
Inside a Group:
Trinity Phase II Network Configuration: • Global link: • 37.6 GB/s • Local link: • Intra-chassis, 5.25GB/s • Inter-chassis, 15.75GB/s (3 tiles). • Intra-Blade link: • PCIE 3.0, 16GB/s *All bi-directional
General: • All-to-all pattern interconnections among routers. • Optical Cable Link among groups. In Trinity, each group link is made with 2 cables. Each cable provides 4.7GB/s bandwidth. • On the cabinet level, there are 2 cabinets in each group, which are connected by backplane electrical links. Each cabinet contains 3 chassis. The bandwidth of each inter-chassis link is 15.75GB/s. • Inside a chassis, the bandwidth of link between each router is 5.25GB/s.
Connections-Router • All nodes connected by routers • 10 inter-group ports • 15 inter-chassis ports • 15 inter-blade ports
Connections-Chassis • 16 blades • 40 connectors to other groups • 5 connectors to other chassis per blade • Backplane connections among blades • PCIE -3 x 16 between a node and blade
Connections-Inter-group • Connection between 2 group ports of 2 routers in 2 groups. • One link between each group • Use Absolute(Direct) pattern.
Datawarp • Burst buffers are implemented as Datawarp nodes in Cray XC40.
Simulation Detail • 96 routers • 10 Burst Buffer nodes • 2 LNET nodes • Final phase 224 LNET nodes • 360 Compute Nodes • 384 Nodes in total, 372 in simulation. • Trinity Phase II, 23 Groups • 230 Burst Buffers • 8280 Compute Nodes • Adaptive routing
Simulation Framework • Application Layer • IOR workload • Darshan 3.1 workload • Model-net Layer • Burst Buffer process • Codes-0.5.2
Results • N-N Write • 4 procs per node • 8MB stripe size on BB • <1024: 32GB per proc • >=1024: 2GB per proc
Results • N-1 Write
Results • N-N Read
Problems • Darshan traces of applications. • Mostly checkpointing at LANL • Lustre is fast enough. • Must use Datawarp APIs. • Datawarp software is still updating. • Currently modeling Trinity Phase II. Final phase is undergoing: • 576 Burst Buffer Nodes • ~20,000 Compute nodes • More modeling details need to be confirmed. • READ bug: • Simulation ends when compute nodes scale up to 2048 in N-N Read. • Ends sooner in N-1 Read simulation. • Seems to be messages are already freed when a reverse event is received.

Recommend

Evaluating Performance of Burst Buffer Models for Real-Application Workloads in HPC Systems

Evaluating Performance of Burst Buffer Models for Real-Application Workloads in HPC Systems Harsh Khetawat Frank Mueller Christopher Zimmer Introduction Existing storage systems becoming bottleneck Solution: burst buffers Use

426 views • 6 slides

Goals Understand use of Dragonfly from game Dragonfly programmers perspective Mostly,

2/7/2012 Goals Understand use of Dragonfly from game Dragonfly programmers perspective Mostly, Project 1 Provide overview of Dragonfly architecture Class diagrams Discuss details needed to fully implement Dragonfly classes

900 views • 42 slides

Performance Advantages of Using a Burst Buffer for Scientific Workflows Andrey Ovsyannikov NERSC,

Performance Advantages of Using a Burst Buffer for Scientific Workflows Andrey Ovsyannikov NERSC, Lawrence Berkeley National Laboratory with David Trebotich, Brian Van Straalen (ANAG, LBNL) BASCD-2016: Bay Area Scientific Computing Day December

582 views • 31 slides

Integration of Burst Buffer in High- level Parallel IO Library for Exa- scale Computing Era SC

Integration of Burst Buffer in High- level Parallel IO Library for Exa- scale Computing Era SC 2018 PDSW workshop Kaiyuan Hou, Reda Al-Bahrani, Esteban Rangel, Ankit Agrawal, Robert Latham, Robert Ross, Alok Choudhary, and Wei-keng Liao

479 views • 19 slides

Accelerating Science with the NERSC Burst Buffer Debbie Bard Big Data Architect, Data and

Accelerating Science with the NERSC Burst Buffer Debbie Bard Big Data Architect, Data and Analytics Services NERSC, LBL July 22, 2016 - 1 - Outline Future compu3ng architecture

754 views • 50 slides

SuperNova burst buffer ( NVMe from Zynq ) Roy Wastie University of Oxford 1 17/10/19 DUNE-UK

SuperNova burst buffer ( NVMe from Zynq ) Roy Wastie University of Oxford 1 17/10/19 DUNE-UK DAQ workshop Specifcation SSD NVMe on the PS of the ZYNQ PCIe Gen2 (4 lanes). Data transfer rate @ 1.6GB/s for >50GB (100s). Data

423 views • 9 slides

TinyOS Determine when Fill message Specify Pass buffer message buffer Network Communication

Inter-Node Communication General idea: Sender: TinyOS Determine when Fill message Specify Pass buffer message buffer Network Communication buffer with data Recipients to OS can be reused Computer Network Receiver:

176 views • 3 slides

Cluster Burst Synchronization in A Scale-Free Network of Inhibitory Bursting Neurons S.-Y. Kim

Cluster Burst Synchronization in A Scale-Free Network of Inhibitory Bursting Neurons S.-Y. Kim and W. Lim Institute for Computational Neuroscience Daegu National University of Education Burst Synchronization - Bursting: Neuronal activity

238 views • 7 slides

Stochastic Burst Synchronization in A Scale-Free Neural Network with Spike-Timing-Dependent

Stochastic Burst Synchronization in A Scale-Free Neural Network with Spike-Timing-Dependent Plasticity S.-Y. Kim and W. Lim Institute for Computational Neuroscience Daegu National University of Education Synaptic Plasticity Stochastic Burst

478 views • 9 slides

Introduction Buffer Overflows Buffer overflows were the most common form of security

Introduction Buffer Overflows Buffer overflows were the most common form of security vulnerability in the 90s. Buffer overflows dominate in the area of remote network penetration vulnerabilities. CS 465 They represent one of the

651 views • 4 slides

Buffer Overflows with Content 2 A Process Stack Buffer Overflow Common Techniques employed

1 Buffer Overflows with Content 2 A Process Stack Buffer Overflow Common Techniques employed in buffer overflow exploits to create backdoors Execution of additional network services via the INETD daemon The addition of new users

497 views • 30 slides

Threads and DragonFly BSD Improving Thread Performance on DragonFly BSD Conduits for program

Threads and DragonFly BSD Improving Thread Performance on DragonFly BSD Conduits for program execution Concurrency A property that allows several vessels of execution to be run without a predefined order. Parallelism A property that

625 views • 23 slides

Observation of THz CSR Observation of THz CSR Burst at UVSOR- -II II Burst at UVSOR 1 Miho

Observation of THz CSR Observation of THz CSR Burst at UVSOR- -II II Burst at UVSOR 1 Miho SHIMADA, 1,2 Masahiro KATOH, 1* Akira MOCHIHASHI, 1 Shin-ichi KIMURA, 2 Masahito HOSAKA, 2 Yoshifumi TAKASHIMA, 3 Toshiharu TAKAHASHI 1 UVSOR Institute

282 views • 25 slides

Scientific Workflows at DataWarp-Speed: Accelerated Data-Intensive Science using NERSCs Burst

Scientific Workflows at DataWarp-Speed: Accelerated Data-Intensive Science using NERSCs Burst Buffer Andrey Ovsyannikov 1 ,Melissa Romanus 2 , Brian Van Straalen 1 , David Trebotich 1 , Gunther Weber 1,3 1 Lawrence Berkeley National Laboratory

306 views • 30 slides

Status of GEO burst analysis efforts Ik Siong Heng for the GEO burst group Outline

Status of GEO burst analysis efforts Ik Siong Heng for the GEO burst group Outline Introduction - who we are - GEOs S3 run S3 analysis - goal: to have list of GEO S3 burst triggers available ASAP - have GEO++ implementation of

401 views • 14 slides

Dragonblood: Analyzing the Dragonfly Handshake of WPA3 and EAP-pwd Mathy Vanhoef and Eyal Ronen

Dragonblood: Analyzing the Dragonfly Handshake of WPA3 and EAP-pwd Mathy Vanhoef and Eyal Ronen ANRW. Montreal, Canada, 22 July 2019. Background: Dragonfly in WPA3 = Password Authenticated Key Exchange (PAKE) Negotiate Provide mutual session

1.04k views • 45 slides

Adaptive Routing in Dragonfly Networks Peyman Faizian , ShafayatRahman Scott Pakin AtiqulMollah,

Traffic Pattern-based C Adaptive Routing in Dragonfly Networks Peyman Faizian , ShafayatRahman Scott Pakin AtiqulMollah, Xin Yuan Mike Lang Florida State University Los Alamos National Laboratory Motivation Dragonfly has been known as a

491 views • 33 slides

Dragonblood : Weaknesses in WPA3s Dragonfly Handshake Mathy Vanhoef and Eyal Ronen BruCON.

Dragonblood : Weaknesses in WPA3s Dragonfly Handshake Mathy Vanhoef and Eyal Ronen BruCON. Belgium, 11 October 2019. 2 Background: Dragonfly in WPA3 and EAP-pwd = Password Authenticated Key Exchange (PAKE) Negotiate Provide mutual

1.97k views • 56 slides

On the Provable Security of the Dragonfly protocol Jean Lancrenon 1 Marjan krobot 1 1

PAKEs Dragonfly Results Conclusion On the Provable Security of the Dragonfly protocol Jean Lancrenon 1 Marjan krobot 1 1 Interdisciplinary Centre for Security, Reliability and Trust University of Luxembourg ISC 2015 1 / 18 PAKEs

915 views • 65 slides

Raw Water Mains Modelling Prioress Mill WPS Network Overview Operational Incident Burst on

Presented by Daniel Gilbert Raw Water Mains Modelling Prioress Mill WPS Network Overview Operational Incident Burst on 48inch Isolate 48 inch Prioress Mill 8 bar Max Is there enough pressure to reach SOR WPS Sluvad Water Treatment

436 views • 7 slides

Dragonblood : Attacking the Dragonfly Handshake of WPA3 Mathy Vanhoef and Eyal Ronen Black Hat

Dragonblood : Attacking the Dragonfly Handshake of WPA3 Mathy Vanhoef and Eyal Ronen Black Hat USA. Las Vegas, 7 August 2019. Background: Dragonfly in WPA3 and EAP-pwd = Password Authenticated Key Exchange (PAKE) Negotiate Provide mutual

532 views • 50 slides

Optimal Adaptive Feedback Control of a Network Buffer V. Guffens, G. Bastin UCL/CESAME (Belgium)

Optimal Adaptive Feedback Control of a Network Buffer V. Guffens, G. Bastin UCL/CESAME (Belgium) American control conference 2005 Portland, Oregon, USA - Juin 8-10 2005 Optimal Adaptive Feedback Control of a Network Buffer p.1/19

511 views • 38 slides

Buffer Manager Each relation is can be mapped onto many files (each file containing data from one

Buffer Management The database buffer is the mediator between the basic file system and the tuple-oriented file system. The buffer managers task is to make the pages addressable in main memory and to coordinate the writing of pages to disk

1.12k views • 8 slides

Discrete Buffer and Wire Sizing for Discrete Buffer and Wire Sizing for Link-Based Non-Tree Clock

Discrete Buffer and Wire Sizing for Discrete Buffer and Wire Sizing for Link-Based Non-Tree Clock Network Link-Based Non-Tree Clock Network Link Based Non Tree Clock Network Link Based Non Tree Clock Network Rupak Samanta, Jiang Hu and Peng Li

713 views • 30 slides