Jetstream: Accessible cloud computing for the national science and engineering communities

funded by the National Science FoundationAward #ACI-1445604

Jetstream: Accessible cloud computing for the national science and engineering communities Matthew Vaughn(@mattdotvaughn)ORCID 0000-0002-1384-4283Director, Life Science ComputingTexas Advanced Computing CenterPI @ Jetstream | Cyverse | Araport | CODE@TACC

“Cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of

configurable computing resources (e.g., networks, servers, storage, applications

and services) that can be rapidly provisioned and released with minimal

management.” NIST Definition of Cloud Computing

http://www.nist.gov/itl/csd/cloud-102511.cfm

SCIENTIFIC COMPUTING THEN

• C/C++/FORTRAN/PERL/SHELL• MPI• LAPACK/BLAS/PETSC• GRID ENGINE• UNIX*• X86/PPC/SPARC• COPROCESSORS

SCIENTIFIC COMPUTING NOWLANGUAGES

• Python 2 & 3• R• Julia• Perl• Matlab• Java• Scala, Clojure, etc• .NET • C/C++• Swift• Haskell• Go• Javascript

FRAMEWORKS

• MapReduce Hadoop, Storm, Pachyderm, Cloudera• Event & Streaming: Kinesis, Azure Stream Analytics, Camel, Streambase• Deep/Machine Learning: Watson, Azure BI, Tensorflow, Caffe• In-memory parsing: Kognito, Apache Spark• Containers: Docker, Rocket, MESOS, Kubernetes• Cloud: AWS, GCE, OpenStack, vCloud, Azure

HARDWARE

•Many-core computing - 50-100 threads/node*

• Xeon / Xeon Phi• GPU• OpenPower• ARM• ShenWei• Google TPU

• Multi-level memory architecture• Hierarchical storage• FPGAs• Quantum-like systems

Funded by the National Science FoundationAward #ACI-1445604

http://jetstream-cloud.org/

What is Jetstream?A production cloud platform for NSF-sponsored researchers

• Provides on-demand interactive computing and analysis• Enables configurable environments and architectures• Supports computational reproducibility and sharing• Democratizes access to cloud-native software• Focused on ease of use for all adopters

Expands the community of users who benefit from NSF investment in shared cyberinfrastructure



Leadership class machinesTraditional HPC, HTC systems

Jetstream serves the Long Tail of Science, Engineering, and Education



Jetstream serves the Long Tail of Science, Engineering, and EducationResearchers, developers, and scientists who

– Need between 1 and a few hundred cores• RIGHT NOW• For the foreseeable future• Not forever

– Want to fully customize the OS and configuration for their research computing environment

– Are working with cloud-native applications & workflows– Use interactive mode for their computing & analytics



• Jetstream is also useful to– Science gateway operators

• Run web applications, databases, and services (front-end)• Use it for on-demand provisioned compute capacity

– STEM educators teaching a variety of subjects• Create a reference VM appliance• Provision an entire classroom• Minimize need for local IT support• Onboard students into XSEDE ecosystem

Jetstream serves the Long Tail of Science, Engineering, and Education



21st Century Workforce Development in the Cloud

• Specialized virtual Linux desktops and applications enable research and research education at small colleges and universities– HBCUs (Historically Black Colleges and Universities)– MSIs (Minority Serving Institutions)– Tribal colleges– Higher-education institutions in EPSCoR States

• Democratize access to cloud-native technologies and approaches – no credit card or PO needed

User PerspectivesUser Accomplishments Role

• Learned how to use the shell and how to work with Linux• Mastered using R to develop plots for his manuscript• Published VM to IU Scholarworks to allow reproducible analysis

Laboratory scientist

• Launches an instance and has full sudo access to customize• Developed a software with R and Python library dependencies• She updates it regularly by creating VM image snapshots

Informatics specialist

• Linked several Jetstream instances with Apache Hadoop• Worked with XSEDE ECSS to import existing Amazon image• Built a simple science gateway to allow others to use his tools

Core facility staff



System Overview

Flavor vCPUs RAM Storage Per Nodem.tiny 1 2 20 46

m.small 2 4 40 23

m.medium 6 16 130 7

m.large 10 30 230 4

m.xlarge 22 60 460 2

m.xxlarge 44 120 920 1

VM Host Configuration• Dual Intel E-2680v3 “Haswell”• 24 physical cores/node @ 2.5

GHz (Hyperthreading on)• 128 GB RAM• Dual 1 TB local disks• 10GB dual uplink NIC• Running Centos7+KVM

Hypervisor

Hardware Specifics

CEPH Storage• 20x Dell 730xd per cloud• 2x10Gbs bonded NIC per 730xd• Running CEPH 0.94.5 Hammer• Configured as OpenStack Storage

• Storage is XSEDE-allocated• Implemented on backend as OpenStack Volumes• Each user gets 10 volumes up to 500GB total storage• Exploring object storage as well but that’s in the future

Platform Overview

Atmosphere APIGlobus Auth

Atmo Services XSEDE Services

OpenStack CEPH

Indiana University

OpenStack CEPH

TACC

OpenStack CEPH

Potentially, Others

Web App

Key components: What is OpenStack?

• Free and open source software for creating private and public clouds

• Started in 2010 by Rackspace and NASA, now managed by OpenStack Foundation

• Widespread adoption across industry and public sector

• Modular architecture with a 6-month update cycle

https://www.openstack.org/

Key components: Everything elseComponent Function

Atmosphere Web application + middleware for providing user-friendly provider-agnostic IaaS

Globus Auth Provides powerful identity and access management functions that are easily integrated into web and mobile applications

XSEDE Services Centralized account management, allocation, and reporting

CEPH Distributed object store and file system designed to provide excellent performance, reliability and scalability

Integration with XSEDE via Globus Auth

Atmosphere Web App uses and Globus Auth implements industry-standard Oauth2• Leaves us flexibility on

identity and access• Globus Auth implements

(in beta) password grant Oauth flow, which means Jetstream access is 100% scriptable with your XSEDE credentials

GateOne Web Shell

• Zero install SSH client• Allows tablets or

Windows to easily use the system

• Supports screen-casting and terminal sharing

Can also use native SSH client with SSH keys

Jetstream is programmable cyberinfrastructure

Web Service APIs

• OpenStack - Official and unofficial clients + libs (i.e. boto)

• EC2* – Integration with AWS-specific code • Atmosphere - Alpha. Getting language

libraries “soon”• Preview @

http://docs.atmospherev2.apiary.io/ Finicky

Automation, Orchestration, and Workflow

• Marathon/MESOS• https://www.youtube.com/v/VzZfwHLmcL0

• Docker Machine + Docker Swarm• Kubernetes• Apache Airavata• CloudMan & Elasticluster• OpenStack MagnumConfiguration Management Tools

• Vagrant / Terraform• Chef• Ansible• Puppet

In Dev

OpenStack API client

Jetstream is like any other OpenStack cloud

• Scriptable access to low-level APIs

• Use language libraries (Python boto, etc.)

• Works* with 3rd party configuration and orchestration tools

openstack image list “show all VM images on this cloud”

Example: Elasticluster

Status: In Development

elasticluster start slurm-js-iu

“Launch a 3-node SLURM cluster with Gluster storage”

Cluster in a box• Slurm, SGE, Torque• Centos, Ubuntu• Hadoop• Gluster, Ceph, NFS• Ansible

-> XSEDE XNIT• compatible clusters

Example: Marathon, MESOS, Docker, Unidata



What comes next?

• Passed acceptance (handily) in May• OpenStack clouds, CEPH storage, and software components functional

– Substantial software integration and development required post-acceptance. Terra incognita!

• Running in “Early operations mode”– Extra maintenance days– A few rough edges (especially for power users)

• Full production - Sep 1, 2016• More features and capabilities will come• Public roadmap soon

Free tier makes it really easy to get started on public cloud$

Our idea: Let any user with active XSEDE User Portal account use a small (but functional) slice of Jetstream• Get an XSEDE account• Sign in to XSEDE User Portal• Click “Trial Jetstream Access”

button• Get access to Jetstream in

about 30 minutes



Jetstream in context

• HTC system• GPGPUs• Virtual Clusters• Gateways

• Data Intensive• Portal-based

configuration• Services• Data sets

• Large Memory• Data Intensive

Computing• Managed VM

• Self-service VM• Gateways• Minimal disk• Extensible

COMET WRANGLER BRIDGES JETSTREAM



Jetstream in context (2)Blue Waters & Stampede 1/2Comet, Bridges, Regional HPC, Public

Cloud$



SummaryJetstream is a public, production cloud platform

• Offers on-demand interactive shell + VNC• Supports configurable software

environments• Enables configurable computing resources• Encourages work with cloud-native software• Empowers novice, intermediate, & expert

users



PartnersConstruction

Application / Community LeadsManagement & Operations

Vendors

[email protected]

@mattdotvaughnwww.slideshare.net/mattdotvaughn

@jetstream_cloud

http://use.jetstream-cloud.org/

http://portal.xsede.org

https://github.com/jetstream-cloud



http://portal.xsede.org/





How can I use Jetstream?• An XSEDE User Portal (XUP) account is required. They are

free! Get one at https://portal.xsede.org• Read the Allocations Overview -

https://portal.xsede.org/allocations-overview• Write a successful allocation request – start with a Startup

or Education request - https://portal.xsede.org/successful-requests

https://portal.xsede.org/

https://portal.xsede.org/

https://portal.xsede.org/allocations-overview

https://portal.xsede.org/allocations-overview

https://portal.xsede.org/successful-requests

https://portal.xsede.org/successful-requests



Where can I get help or learn more?• Production:

– Web app: http://use-jetstream-cloud.org/– User guides: https://portal.xsede.org/jetstream– XSEDE KB: https://portal.xsede.org/knowledge-

base– Email: [email protected]– Campus Champions:

https://www.xsede.org/campus-champions– Training Videos / Virtual Workshops (TBD)



https://portal.xsede.org/jetstream

https://portal.xsede.org/knowledge-base

https://portal.xsede.org/knowledge-base

mailto:[email protected]

https://www.xsede.org/campus-champions

https://www.xsede.org/campus-champions

Expanding NSF XD’s reach and impactAround 299,000 researchers, educators, & learners received NSF support in 2012-2013

– Only 1.5% completed a computation, data analysis, or visualization task on XSDEDE program resources

– Less than 3% had an XSEDE Portal account– 70% of researchers surveyed* claimed to be resource

constrainedWhy don’t they use XSEDE systems?

– Activation energy is pretty high– HPC resources are scarce and not well-matched to their needs– They just don’t need that much capability

* https://www.xsede.org/xsede-nsf-release-cloud-survey-report

Technology

Jetstream: Accessible cloud computing for the national science and engineering communities