Slide Navigation






[[ Slide navigation:

Forwards:right arrow, space-bar or enter key
Reverse:left arrow

]]    




 

Introductions


2018 October RLP Options M&K Workshops

Simon Hood

its-ri-team@manchester.ac.uk

Head of Research IT Infrastructure and Platforms,
IT Services, University of Manchester




 

RLP Intro

RLP Overview

Workstream 1: e-Research Infrastructure
  • Compute and storage — on-site and Cloud;
  • people!
Workstream 2: Research Data Lifecycle/Management
  • Research data storage and archive;
  • publication and respositories.
Workstream 3: Research Administration
  • Grant development, funding management. . .
  • Processes, tools, systems, teams. . .




 

Why are we here today?


To determine how to spend 360k on compute resource

  • What computational resources are missing?
  • What would you like more of?
  • How can we make access to/use of them better?




 

Existing Platforms Overview

Batch Compute
  • CSF3: STD almost 9,000; HPC Pool 4096 cores; 26 * 512 GB nodes;
  • CSF2: 2048 AMD cores (deprecated);
  • Condor: 5,489,796 CPU core-hours in last 12 months.
Interactive Compute
  • iCSF: 10 * 64 GB nodes; 8 * 256 GB nodes; 1 * 2 TB nodes.
VM Platforms
  • RVMS — now dual-site VM platform.
AWS
  • In (very) soft launch. . .




 

The UoM CIR Ecosystem

The University CIR Ecosystem.




 

Phase One M and K


What did we procure with last year's money?

  1. 16 * Nvidia V100 GPUs, 182k.
  2. 3 * 1.5 TB RAM compute nodes, 72k.
  3. 10 * STD compute nodes (320 cores), 72k.
  4. Cloud (AWS) resources, 84k.
    • e.g. for Shiny, Jupyter. . .




Ordered in January. ETA end of April. . .




 

Recent/Upcoming Developments

In Production: RLP 'Z'
Upcoming
  • Soon: 52 Nvidia V100 GPUs available
  • 12 FATPOU; remainder priority to contributors.
AWS via Research IT (RLP 'O')
  • RIT AWS Organisations;
  • enables RIT tech support; direct charges to grant accounts. . .




 

Your Input


Your input required!




 

Possible Investments

Possible investments for the 360k:

  • HTC
  • HPC
  • GPUs
  • high-mem (RAM)
  • Web-fronted
  • Cloud
  • Spot (non-urgent work)
  • on-demand (urgent work)
  • Other. . .