ceph for tier-2 storage at msi...© 2014 regents of the university of minnesota. all rights reserved...

© 2014 Regents of the University of Minnesota. All rights reserved.

Minnesota Supercomputing Institute for Advanced Computational Research

Ceph for Tier-2 Storage at MSI

Benjamin Lynch, Ph.D.

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Campus and External Hosts

High-Performance Computing Clusters

Interactive HPC

2nd Tier Persistent Storage

(Ceph/S3)

High-Performance

Persistent Storage

and Scratch Space

(PANFS)

Tape Backup

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Ceph / S3

MSI Private Cloud

Public Cloud

Web

MSI HPC

Globus

Hadoop

and Spark

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Web

Ceph / S3

MSI HPC

Public Cloud

Globus

Hadoop

and Spark

MSI Private Cloud

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Web

Ceph / S3

MSI Private Cloud

Public Cloud

Globus

MSI HPC

Hadoop

and Spark

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Ceph / S3

MSI HPC

MSI Private Cloud

Public Cloud

Hadoop

and Spark

Web

Globus

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Ceph / S3

MSI Private Cloud

Public Cloud

Web

MSI HPC

Globus

Hadoop

and Spark

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Ceph / S3

MSI Private Cloud

Public Cloud

Web

MSI HPC

Globus

Hadoop

and Spark

Scalable volume

Scalable performance

High availability

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Why Ceph?

•  Scalable •  Fault-tolerant •  Flexible configuration •  Flexible support

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

Our Ceph Cluster •  Cluster Specs

–  Ceph 0.94.5 (Hammer) –  3 PB raw disk –  4+2 erasure coding for most data –  4 load-balanced gateways (civetweb) –  10GigE network –  3 monitors

•  Applications –  Genomics workflows on public & private cloud compute resources –  Data staging for HPC applications –  Data staging for Hadoop / Spark cluster –  Bulk store of stale data

Minnesota Supercomputing Institute for Advanced Computational Research

© 2014 Regents of the University of Minnesota. All rights reserved.

•  Luke Burns (System Administrator) •  Matt Mix (Network Administrator)

Acknowledgements