the marriage of cloud, hpc and containers

17
THE MARRIAGE OF CLOUD, HPC AND CONTAINERS ...AND SERVERLESS?

Upload: inside-bigdatacom

Post on 15-Mar-2018

224 views

Category:

Technology


0 download

TRANSCRIPT

Page 1: The Marriage of Cloud, HPC and Containers

THE MARRIAGE OFCLOUD, HPC AND

CONTAINERS...AND SERVERLESS?

Page 2: The Marriage of Cloud, HPC and Containers

ADAM HUFFMANSenior HPC and Cloud Systems Engineer

The Francis Crick Institute

@adamhuffman

[email protected]

Page 3: The Marriage of Cloud, HPC and Containers

THE FRANCIS CRICK INSTITUTEBiomedical discovery institute dedicated to

understanding the fundamental biology underlyinghealth and disease

Biggest biomedical research facility under a single roofin Europe

1,500 staff, £650m investment

Page 4: The Marriage of Cloud, HPC and Containers

HPC STATUS QUO IN PERILPressure from funding bodies

Pressure from procurement

Deadline-driven bursting

Page 5: The Marriage of Cloud, HPC and Containers

EMEDLAB

Biomedical research cloud, with 6,048 cores and 5.5PBstorage

Share data, methods and expertise

OpenStack, GPFS

http://www.emedlab.ac.uk

Page 6: The Marriage of Cloud, HPC and Containers

GPFSBenign environment assumption

Licensing complications

NFS export

Page 7: The Marriage of Cloud, HPC and Containers

IMPORTANCE OF SHARED FILESYSTEMACCESS

Scientific reference datasets

Object storage?

Application support for objects?

Page 8: The Marriage of Cloud, HPC and Containers

DATA ACCESS EXAMPLE800TB dataset from TCGC

Restricted access

Special client

Multiple instances

Page 9: The Marriage of Cloud, HPC and Containers

VIRTUAL CLUSTERSElasticluster

Heat

Ansible/Salt etc.

Back to importance of shared filesystem access

Are they actually the best fit for clouds?

Page 10: The Marriage of Cloud, HPC and Containers

THE COMFORT OF CLUSTERSHow well are the characteristics of pipelines known?

Only start VMs as needed

Reflects focus on high utilisation, rather than "time toscience"

Page 11: The Marriage of Cloud, HPC and Containers

PERFORMANCE/FLEXIBILITYTRADEOFFS

Performance tuning of clouds

Flexibility cost

Rising so�ware complexity

Page 12: The Marriage of Cloud, HPC and Containers
Page 13: The Marriage of Cloud, HPC and Containers

CONTAINERS THE UNSTOPPABLEHYPE MACHINE

Hyperscale mindset and capabilities

Maturity of container support for scientificapplications

Singularity, Shi�er "fixing" containers for scientific use

Are we just reinventing packaging, badly?

Page 14: The Marriage of Cloud, HPC and Containers

SERVERLESSLambda

Google Cloud Functions

OpenWhisk

Fission

Will science have a say?

Page 15: The Marriage of Cloud, HPC and Containers

ABSTRACTION LAYERSUsers focus on images and snapshots

Is this the right level of abstraction

Use repeatable configurationmanagement/provisioning instead?

Criticism of TripleO for being image-centric

How to update them

Image catalogues

Page 16: The Marriage of Cloud, HPC and Containers

CMS/CERN HIGH ENERGYPHYSICS

Simply treat clouds as nice new resources

Already have tools to handle remote datatransparently

Already have accounting tools

Advantages of grid experience

Concerns about security

Page 17: The Marriage of Cloud, HPC and Containers

THANK YOUImage Credits:

The Guardian

Alvaro Sanchez