SLIDE 1
Query Processing in a Self-Organized Storage System
Hannes Mühleisen, supervised by Robert Tolksdorf
1
- Scalability
- Data, Queries, Nodes
- Robustness
- Node/Network failure
- Adaptiveness
- “Fair” distribution of load
2
Query Processing in a Self-Organized Storage System Hannes - - PowerPoint PPT Presentation
Query Processing in a Self-Organized Storage System Hannes - - PowerPoint PPT Presentation
Query Processing in a Self-Organized Storage System Hannes Mhleisen, supervised by Robert Tolksdorf 1 Distributed DBs - Goals Scalability Data, Queries, Nodes Robustness Node/Network failure Adaptiveness Fair
SLIDE 2
SLIDE 3
S1 S2 S3 S4 S5 S6 C1
Clustered / Federated
3
[Bernstein81, Epstein78]
SLIDE 4
S1 S2 S3 S6 S5 S4
1-2 0-1 2-3 3-4 4-5 5-6
Global Laws
4
[Harren02,Karnstedt04,Rösch05]
SLIDE 5
S1 S2 S3 S6 S5 S4
#B
70% 25% 95% 50% 50% 95% 10% 85%
#B?
Probabilistic Request Routing
5
[Lindgren03]
SLIDE 6
[Wilensky97, NetLogo Ants model]
6
SLIDE 7
Distribution Paradigms
Scalability Adaptability Robustness Completeness Complex Queries Stand-Alone Federated Global-Law Probabilistic e.g. Swarms
low high low high ✓ high high fair high ✓ high fair high high ✓ high high high fair ?
7
SLIDE 8
Research Question
Can complex queries be evaluated efficiently in a swarm-based distributed storage system?
8
SLIDE 9
Mutable Moving Query Plans
parse ✓ rewrite✓ move & repeat
↺
Catalog✗
- ptimize
based on?
↺
execute where?
↺
9
[Papadimos03,Battré08]
SLIDE 10
10
↺
⋈ σ r r r σ
③ ① ②
↺
⋈ σ r r r σ
③ ① ②
↺
⋈ σ r r
③ ②
SLIDE 11
11
↺
r(#)
① ② r(*)
r(*)
① ② r(#)
↺
p(#)= 2% p(*)=78% p(#)= 2% p(*)=10% p(#)=53% p(*)= 3%
SLIDE 12
12
Handling Routing #Failures
r(#)
①
↺
p(#)=0% p(#)=0%
what now? Trackback!
SLIDE 13
13
Evaluation Methodology
3,75 7,5 11,25 15 Query 1 Query 2 Query 3 Query4
# Participating Nodes / Query
Optimal Plan Moving Mutable Plans Static Plan Routing
better
Not actual data!
SLIDE 14
14
Evaluation Methodology
150 300 450 600 Query 1 Query 2 Query 3 Query4
# Results / Query
Optimal Plan Moving Mutable Plans Static Plan Routing
better
Not actual data!
SLIDE 15
Thank You!
Web Page: http://hannes.muehleisen.org