AWS for scientific computing
          jss 2011-05-12
TOC

•   Overview: Cloud Eco System

•   Amazon Web Services for scientific computing

    •   EC2 Deployment

    •   S3, EBS and local Storage

    •   Queueing / Messaging
Cloud Layers


Software as a Service - SaaS
   running applications
Platform as a Service - PaaS
    software environments and services
Infrastructure as a Service - IaaS
    virtual machines and network
Elastic Compute Cloud (EC2)

•   Virtual Data Center

•   Instance Types

•   Regions

•   Availability Zones

•   Security Groups (Firewalling)
                                    Can go down just like anyone else
•   Amazon Machine Images (AMI)
EC2 Setup

•   Management:
    •   AWS web console
    •   CLI: ec2-tools
    •   Python: boto (check dev/src/tba)

•   Use the setup script

•   Get SSH fingerprint from instance’s boot console

•   Use some form of parallel SSH for additional setup (Twisted conch
    works great, fabric for serial deployment)
The DNS Problem

•   Instances get new IP on boot

•   Solutions:

    •   use Virtual Private Cloud (Amazon VPC)

    •   auto-update hosts file

    •   set up DNS server
Storage


•   Instance
    fast, cleared on boot, no charges

•   Elastic Block Storage (EBS)
    fast, networked, instant snapshot-able, N attachments per instance

•   Simple Storage Service (S3)
    HTTP access, CDN (CloudFront)
Messaging


•   Simple Queueing Service (Amazon SQS)

•   AMQP / RabbitMQ (c.f. knk’s recent rocketcircus)

•   Rolling your own (e.g. XML RPC) is fine, too
TLDR


AWS for the occasional number crunch:

      •   1/2 day to get basic setup running
          (longer for robust setups)

      •   profit

AWS Quick Intro

  • 1.
    AWS for scientificcomputing jss 2011-05-12
  • 2.
    TOC • Overview: Cloud Eco System • Amazon Web Services for scientific computing • EC2 Deployment • S3, EBS and local Storage • Queueing / Messaging
  • 3.
    Cloud Layers Software asa Service - SaaS running applications Platform as a Service - PaaS software environments and services Infrastructure as a Service - IaaS virtual machines and network
  • 4.
    Elastic Compute Cloud(EC2) • Virtual Data Center • Instance Types • Regions • Availability Zones • Security Groups (Firewalling) Can go down just like anyone else • Amazon Machine Images (AMI)
  • 5.
    EC2 Setup • Management: • AWS web console • CLI: ec2-tools • Python: boto (check dev/src/tba) • Use the setup script • Get SSH fingerprint from instance’s boot console • Use some form of parallel SSH for additional setup (Twisted conch works great, fabric for serial deployment)
  • 6.
    The DNS Problem • Instances get new IP on boot • Solutions: • use Virtual Private Cloud (Amazon VPC) • auto-update hosts file • set up DNS server
  • 7.
    Storage • Instance fast, cleared on boot, no charges • Elastic Block Storage (EBS) fast, networked, instant snapshot-able, N attachments per instance • Simple Storage Service (S3) HTTP access, CDN (CloudFront)
  • 8.
    Messaging • Simple Queueing Service (Amazon SQS) • AMQP / RabbitMQ (c.f. knk’s recent rocketcircus) • Rolling your own (e.g. XML RPC) is fine, too
  • 9.
    TLDR AWS for theoccasional number crunch: • 1/2 day to get basic setup running (longer for robust setups) • profit