HPC Solutions

HPC Solutions

As with all technology solutions, it is imperative to identify the requirements, it will be necessary to work the with application vendors and teams to determine the specific HPC requirements. This can be done via workshops and also using HPC benchmarking tools.

HPC metrics:

  • Scale-out or Scale-up
  • Memory, CPU (NUMA) or Disk (IOPS) Intensive
  • GPU or CPU intensive
  • 64, 32 or multi or single tread applications or
  • Batch Process or Cluster
  • File/Block/SMB/NFS/CIFS
  • OS
  • Application
  1. What is your Current Server Network Connectivity Media and Protocol?
  2. What is your current OS and Application?
  3. What is your Total CPU Ghz requirements?
  4. What is your IOPS and (Read or Write) intensive?
  5. What is your total memory required?
  6. What is the Total storage Required?
  7. What is the your Switch Fabric?
  8. Do you have any specific requirements for NUMA, GPU or Intel Phi?
  9. Is your application single thread or multi-thread?
  10. Do you use a Bench Marking Tool? (LinPack)
  11. Do you require File or Block?
  12. Do you require Scale-Out or Scale-Up?
  13. How do you load Data?
  14. How do end users access the Application?
  15. What is the speed of your Network and Storage?

Generic HPC technologies:

  • Intel Phi
  • nVidia Grid
  • Intel / AMD CPUs
  • IBM iDataplex – Most common servers type for generic clusters
  • InfiBand
  • 10 GB
  • PCIe Flash
  • Sold State Storage Systems
  • Scale out Storage
  • Converged Infrastructure
  • Software Defined Storage and Networking
  • Parallel File Systems
  • hierarchical storage management (HSM)
  • ASIC Based Switches
  • NUMA
  • Low Latency Switches
  • 40/56GbE Ethernet RDMA over Ethernet

  • Low Latency Switch (Cisco, Arista, Mellanox – SX6036)
  • Low Latency NICs (Mellanox InfiBand Over IP) – 10 GB
  • Low Latency FC,IP 10 GB, FCoIP,NFS, SMB, IB = 30 Gb/s
  • Storage (Isilon,XtremeIO)
  • Compute (SuperMicro FatTwin + Intel Phi + nVidia GRID + XtremeCache + XtremeFS)

Process

  1. Workshop technology options
  2. Identify Application requirements and IOPS and document success criteria
  3. Benchmark existing platform (if any)
  4. Short list vendor solutions for Compute, Storage, Networking, Cloud
  5. Shortlist Vendor presentations and evaluation
  6. High Level design and Costs
  7. PoC/Prove Benchmarks
  8. Implement

High level process for HPC workshop:

  • Workshop technology options and high level requirements and plan
  • Identify Application requirements and IOPS and document success criteria
  • Benchmark existing platform
  • Shortlist vendor solutions for Compute, Storage, Networking, Cloud
  • Shortlist Vendor presentations and evaluation
  • High Level design and Costs
  • PoC/Prove Benchmarks
  • Implement

HPC Bbenchmark

  • linpack

Traditional HPC Options

  • Intel Phi
  • nVidia Grid
  • AMD Opetron
  • AMD SeaMicro
  • IBM iDataplex – Most common servers type for generic clusters
  • HP Moonshot
  • HP Apollo
  • InfiBand
  • Flash PCIe
  • Arista, Mellaonox
  • Fibre Switches
  • 10 GB Ethernet
  • SD Networking
  • SD Storage
  • Solid State
  • Quantum
  • Virtualise Data
  • Virtualise OS
  • Nutanix
  • WebScale IT
  • panasas storage
  • http://www.arista.com/en/products/7500-series
  • Converged
  • UCS
  • http://www.sgi.com/products/servers/

Research

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s