Distro compute

616 views

Published on

Distributed Compute in Windows Azure, prepared for ACCU London

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
616
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • Developer centres forJavaPHPRubyPython
  • Took 1.9 million compute hoursWould have taken 25 years on an 8 core machine
  • Cost vs recovery time
  • You can’t scale a HPC cluster as it can’t open up new connections
  • Scalable clusters are great but better is a flexible cluster that you can change dynamically at runtime
  • If the system is totally torn down at the end, you lose all sense of logs.
  • Distro compute

    1. 1. Introduce
    2. 2. Workloads Distributed compute in a nutshell (where many nuts > few nuts)
    3. 3. Workloads
    4. 4. Distributed Compute Developer Big Compute Big Penguin Big Data
    5. 5. BIG ANYTHING Hello All Worlds
    6. 6. Big Co$t? That’d be like Microsoft, right?
    7. 7. Microsoft Research Genomics
    8. 8. Inspect Distributed compute in a nutshell (with many little nuts)
    9. 9. Inspect
    10. 10. HPC Head Node Broker Nodes Compute Nodes Allows on-premises And hybrid option Compare Architectures Big Data Name Node Data Nodes Allows cloud or on-premises no hybrid option
    11. 11. Hadoop HPC
    12. 12. All distributed compute works on the basis of taking a large JOB and breaking it to many smaller TASKS which are then parallelised
    13. 13. Develop
    14. 14. Deploy
    15. 15. Examples How do you take your compute?
    16. 16. OUCH Lessons learned from getting too close to the coalface
    17. 17. A broken cluster is no place to be diagnosing
    18. 18. Scalability < Elasticity
    19. 19. Hybrid HPC is next to useless
    20. 20. 95% through a petabyte is a bad place to find a bug
    21. 21. Thank you!

    ×