ingeneos: intercontinental genetic sequencing over trans-pacific networks and supercomputers

13
nci.org.au nci.org.au @NCInews InGeneoS Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers Jakub Chrzeszczyk, NCI Cloud Team Andrew Howard, NCI HPC Team

Upload: andrew-howard

Post on 18-Jul-2015

33 views

Category:

Documents


4 download

TRANSCRIPT

Page 1: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.aunci.org.au

@NCInews

InGeneoS Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

• Jakub Chrzeszczyk, NCI Cloud Team • Andrew Howard, NCI HPC Team

Page 2: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

InGeneoS Components

• NCI and A*Star collaboration • Goals

• Utilise trans-Pacific extended InfiniBand and shared SuperComputer resources to accelerate DNA analysis • Transfer large (~300Gb) genetic sequence data

sets generated in Canberra from NCI to A*Star Singapore for analysis on the A*Star Aurora large memory system with the results visualised in New Orleans (SC14)

• Utilise NCI High Performance InfiniBand Cloud HPC systems for visualisation of genetic data results produced by Aurora

Page 3: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Genetic Sequence Workflow for SC14

GenETIC SEQUENCE: GATACGGAGTTTA………A 381GB

NCI

AUA*Star

SG

381G

1143G

Page 4: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Network

Singapore/Canberra10Gbs (~30,000km)

Canberra/Singapore1Gbs (~6000km)

Page 5: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Extended InfiniBand Fabric

Obsidian E100InfiniBand Range Extender

Obsidian E100InfiniBand Range Extender

Obsidian E100

NCI

AU

Obsidian E100InfiniBand Range Extender

Obsidian E100InfiniBand Range Extender

TITechJP

Obsidian E100

Obsidian E100

Obsidian Crossbow

InfiniBand Switch

Obsidian E100InfiniBand Range Extender

GaTechUS

SC14US

A*StarSG

Page 6: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Experiment 1: AU to SG large data transfer, process in SG on Aurora, return results data to AU

A*Star Aurora

Obsidian E100InfiniBand Range Extender

CX250CX250

Obsidian E100InfiniBand Range Extender

301Gb

301Gb

SingAREN

SG

NCI

Mellanox 6036InfiniBand Switch

AARNet

AU

PNWGPUS

CX250

CX250Aurora

ProcessResultData

NCICanberraAU

A*Star

SG

LustreFile

System

Page 7: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

NCI InfiniCloud: HPC InfiniBand performance in a Cloud

NCI InfiniCloud

InfiniBand

LustreFile

System

OpenStack

LustreFile

System

Page 8: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

NCI InfiniCloud: HPC InfiniBand performance in the Cloud

• OpenStack Cloud supported by a 56Gbs InfiniBand fabric • High Performance IB MPI • High Performance Lustre File System access

• Built using Mellanox Neutron modules • Flexible top performance computational

resources ‘on demand’ • Flexible OS and application stack supported by

an I/O architecture which includes local Solid State Disk and high performance large file storage using Lustre

Page 9: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Experiment 2: SG to AU data transfer, process on InfiniCloud, view results at SC14

SC14US

A*Star

SG

Obsidian E100InfiniBand Range Extender

CX250

CX250

CX250

CX250

Mellanox 6036InfiniBand Switch

Obsidian E100InfiniBand Range Extender

Obsidian E100InfiniBand Range Extender

Data

SingAREN

SG

PNWGPUS

AARNet

AU

NCIInfiniCloud

Obsidian E100InfiniBand Range Extender

Process

Display

ESNETUS

Page 10: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

AU to SG data transfer speedAv

erag

e Ti

me

0h

1h

2h

3h

4h

5h

298Gb data set rsync NCI Io A*Star R&E path rsync NCI to A*Star 10Gbs IPoIB DSYNC 1Gbs IB path DSYNC 10Gbs IB path (~26,000Km)

7m

1h 21m

3h

4h 7m

Page 11: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

SG to AU data transfer speed Av

erag

e Ti

me

0h1h2h3h4h5h6h7h8h9h

10h11h12h13h14h

1,143Gb data set rsync A*Star to NCI R&E path rsync A*Star to NCI 10Gbs IPoIB DSYNC 1Gbs IB path DSYNC 10Gbs IB path (~26,000Km)

24m

4h 5m

9h

12h 33m

Page 12: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

SG to AU Observed data transfer rate at 10Gb using Obsidian DSYNC

1 second 900 MB

1 minute 54 GB

1 hour 3.24 TB

1 day 77 TB

1 week 539 TB

1 month 2.3 PB

Page 13: inGeneoS: Intercontinental Genetic sequencing over trans-Pacific networks and Supercomputers

nci.org.au

Thanks

This project made possible with the kind assistance of

A*Star ANU John Curtin School of Medical Research Obsidian Strategics SingAREN AARNet Pacific Northwest GigaPOP ESNet

and anyone else we’ve forgotten to thank