2008-09-16

From RAD Lab

Jump to: navigation, search

Contents

Attendees: Dave Patterson, Armando Fox, Mike Franklin, Michael Armbrust, Beth Trushkowsky, Jesse Trutna, Nick Lanham, Haruki Oh

[edit]

Whiteboard Notes

[edit]

Not New

  • Multiple Indices
  • Eventual Consistency
  • Scale
[edit]

New

  • Arbitrary Index Functions
  • Quantifiable
  • Consistency
  • Declarative Performance Spec
    • Need Examples
  • Indexes for everything?
  • More Efficient Archiving
    • save logical updates 
We should get feedback on what Mike Franklin meant by "arbitrary" indexes..
[edit]

Possibly Interesting

  • Utility Computing
    • Rapid Scale Down
      • Heterogeneous Nodes 
Heterogenous Nodes := Disk-heavy persistent nodes, 
dynamically allocated memory-heavy processing nodes
[edit]

Plausible prototype plan

  • EC2/S3?/EBS
  • Cassandra
  • Cloud Stone
    • scale up/down
  • kill virtual machines 
Discussion surrounding S3 usage as archive medium
(since we don't have persistence guarantees from EC2)
with possibility of bootstrapping new nodes from
archive instead of other processing nodes, since
they're likely busy and S3 has high bandwidth.
[edit]

To Do

  • DBs and Virtualization
  • Categorize Workload
    • How different from traditional data intensive apps (BigBank, etc)
  • Articulate why databases don't scale
    • CAP
  • Back of Envelope Calculation
    • EC2/S3/EBS

Delayed Charguwww while indexes update 

CAP := Consistency, Availability, Partitioning
CAP Theorem states you can't have all three.
Traditional databases focus on Consistency (via ACID)
[edit]

Whiteboard

scan of whiteboard from meeting
Enlarge
scan of whiteboard from meeting
Retrieved from "http://radlab.cs.berkeley.edu/wiki/2008-09-16"