Cloud Computing Panel

Hook

• Key points: scaling, spot market but need resiliency(99.5% for 2000 nodes -> some nodes down per month), fault tolerant

• Optimizing S3 Object Store Performance

  • S3 object key naming affects performance -> object partition

• Market Maker:

• Data lake: collocation; minimize data movement; maximize user services

  • hot data lake: long-term public cloud storage is expensive; object store data for hot data.

Mike Little:

• The role of AIST: advanced information technologies for Earth science space-based and ground-based information systems; accessibility and utility of science data

• Two Major Cloud Efforts: AMCE; Technology Developments which need cloud computing (SAR Science data processing: iISCE, ARIA series of projects)

• AMCE: Overview

  • enable PI-led teams access to as much capability as possible

  • avoid financial management and procurement problems

  • minimize computer security problems, and management overhead costs

  • reduce cost of refactoring (OS, libraries, security, central authority)

  • avoid re-installation by SaaS

• SAR SDP: various data products, science data processing(high volume, data quality, cloud with scalability)

Chris

• Archive Volume increase fastly

• EOSDIS Cloud Plans:

  • Object storage

  • Cloud Analytics Prototypes

• Big questions:

  • how to supply data to all users on a non-discriminatory basis?

  • How to avoid storage lock-in?

  • How to attract users to cloud?

  • How to predict pricing?

  • How to migrate near archive data to cloud?

  • What should not go to cloud?

  • How to handle provisions and accounting of cycles?

  • Do we need new operations policies or procedures?

Wilson

• advantages: Replicating services for High availability or Backup

• Parallel Computing: Multi to many core; GPU daughter boards; compute clusters (MPI, OpenMP; Batch ma-reduce; in-memory map-reduce); Cloud computing

• Comparison of Big Data Platforms (“A Survey on Platforms for Big Data Analytics”)

• Data Locality/Movement Challenge

  • the data communication in cores, gpu, cpu, cloud

• Cloud or Not?

  • Cloud compute VS. Buying hardware

  • Serving data and archive

  • Security

  • Modular, high-availability

Question:

• Storage lock-in, what is the data policing?

  • relatively: some are big data; some are not;

• Comment on cloud on education?

  • NASA cloud training;  

• Data accessibility in cloud?

  • on-demand access

• the capability of the cloud computing, is it the future direction?

  • not sure.

• Risk of achieve data

  • one of biggest worrying; build own data centers;

  • public cloud & hybrid cloud

• How to move the data to compute?

  • most cases are moving computing to data.

  • multi-copying to improve performance