coming to grips with the power proportional data storage problem n.
Download
Skip this Video
Loading SlideShow in 5 Seconds..
Coming to Grips with the Power Proportional (Data) Storage Problem PowerPoint Presentation
Download Presentation
Coming to Grips with the Power Proportional (Data) Storage Problem

Loading in 2 Seconds...

play fullscreen
1 / 7

Coming to Grips with the Power Proportional (Data) Storage Problem - PowerPoint PPT Presentation


  • 115 Views
  • Uploaded on

Coming to Grips with the Power Proportional (Data) Storage Problem . Sara Alspaugh and Arka Bhattacharya. State of The Art : case 1 solved, case 2 solved in many instances - yet unclear if done in practice case 3 open. What : power proportional storage storage paradigms:

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha
Download Presentation

PowerPoint Slideshow about 'Coming to Grips with the Power Proportional (Data) Storage Problem' - didina


An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
coming to grips with the power proportional data storage problem

Coming to Grips with the Power Proportional (Data) Storage Problem

Sara Alspaugh and Arka Bhattacharya

slide2

State of The Art:

  • case 1 solved, case 2 solved in many instances
    • - yet unclear if done in practice
  • case 3 open

What:

  • power proportional storage
  • storage paradigms:

case 1: static, replicated

case 2: monolithic, dedicated

case 3: co-located with computation

Results:

New Ideas:

  • per-rack SSD cache (case 3)
  • power-proportional structured storage contributions (case 2)

more good

less good

1 web farms
1 . Web Farms
  • Examples: most websites
  • State of the art: easily made power proportional (Chen [SIGMETRICS ‘05], NapSac [SIGCOMM GreenNets ‘09], etc.)
    • mostly static, replicated content , serving identical requests
2 monolithic storage tier
2 . Monolithic Storage Tier
  • Examples: search (in-memory indexes), email (disk), etc. – CFS, SAN, transaction tier
  • State of the art: power proportional distributed file systems (Sierra [MSR-TR ‘09], Rabbit [SOCC ‘10]) and power proportional SAN/RAID arrays (Hibernator [SOSP ‘05], etc.)
  • Opportunities in structured storage
  • Trade-off load balancing, replication, fault-tolerance, read performance versus consistency, write performance, power
    • exactly how depends on storage model and level of abstraction
3 distributed storage and computation co located
3 . Distributed Storage and Computation Co-located
  • Examples: DFS + data parallel runtime (Cosmos + Dryad, HDFS + Hadoop)
  • State oftheart: FAWN [SOSP ‘09]
  • Same trade-offs as previous case
  • Other considerations:
    • per-rack SSD cache
    • what if data-locality is not important?
case study results
Case Study / Results
  • power proportional key-value stores and friends
  • knobs:
    • metadata (centralized or decentralized)
    • degree of replication
    • consistency model
    • workload model
    • service level objective
    • cost (hardware and electricity)

cost

power savings

power savings

latency / power savings

power savings

read latency

write latency (strict consistency)

write latency (eventual consistency)

workload locality

replication