Arches MetaCluster

1
Arches MetaCluster

• Brian Haymore
• Eli Stair
• Ron Price
• http//www.chpc.utah.edu/general/presentations/Met
  aCluster

2
Overview

• From Beowulf to MetaCluster
• Where did we come from?
• Where are we at?
• Where do we hope to go?

3
Where did we come from?

• Beowulf mixing to many disparate resources
  (memory, CPU speed, architecture, interconnect)
• Problem causes overly complex policies
• Mixing the wrong things fewer more expensive
  resources that arent used efficiently.
• Shotgun Effect

4
Where did we come from Goals created from past
experiences

• Automated infrastructure to provide a user with a
  more convenient interface to resources.
• Lessons learned putting mms and skittles in one
  bucket doesnt make it easier to get what you
  want.
• No more Shotguns, need the silver bullet Buy
  hardware according to need.

5
Where are we at?

• FLYER IMAGE

6
Where are we at? (cont)

• FLYER IMAGE PLUS cluster names and notes on how
  we avoided the shotgun effect and fixed other
  things

7
Were are we at

• By building task focused resources we optimized
  our cost. We didnt have to buy every feature
  inside our box. Biggest need is low latency
  followed by????

8

• Low latency delicate, landscape
• High mem tunnel
• Viz skyline
• Marching men general cycles

9
Were Do We Hope to Go?

• Less general purpose and tailored our resources
  for applications with special needs.
• Pic from flyer plus Globus and interfacing with
  other sights

10
Misc

• The Future of Grid
• Give and be part of the Grid community
• Single sign on
• Allows us to span political entities
• Look into NWS
• Minimize the different syntax and policies of
  various PBS installations/configuration

11

• Do demo, GET WATER, get time lapses
• Know what two days we will target, avoid key
  notes,

12
Question Session

• Feel free to email me a question
  rprice_at_chpc.utah.edu
• Slides on the web at
• http//www.chpc.utah.edu/rprice/presentations/G
  rid_Computing_and Bioinformatics.pdf