Daily and weekly patterns Workload classes: working days, holidays, outliers Trends Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 3 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Benchmark Design Challenges Design Challenges Realistic data & workload patterns Scalable Configurable Random generators
Basis Stud.IP eLearning management system Online application 1 year web server log data Complete database dump Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 4 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Benchmark Design outbox user_id message_id
messages message_id timestamp sender_id receiver_id subject content studiengaenge studiengang_id name
30 queries
institute institute_id name address
Only course management Most important: seminar - seminar user - user
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 5 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Data Generation Reference Distribution of Users per Seminar in Table seminar_user 4500 User References
4000
Log−Normal Distribution
Based on original data
3500
3000
Entries
Maximum likelihood estimation Standard probability distributions
2500
2000
1500
1000
Scalable
500
0
Reference distribution in table seminar_user
10
20
30
40
50 60 Users per Seminar
70
80
90
100
Reference Distribution of Seminars per User in Table seminar_user 4500 Seminar References Log−Normal Distribution Gamma Distribution
4000
100 3500
60
3000
40
2500
Entries
#Entries
80
20 0 100
2000
1500
100
80 80
60
60
40
1000
40
20 #Seminars per User
20 0
0
#Users per Seminar
500
0
10
20
30
40
50 60 Seminars per User
70
80
90
100
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 6 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Workload Modeling Workload interpreted as time series Classes of queries (meine seminare) Classes of days (Monday during lecture period) Daily approximation by polynomials June 1 − July 6, 2008 12000 plugins extern seminar_main meine_seminare
10000
Page Hits
8000
6000
4000
2000
0 0
1
2
Weeks
3
4
5
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 7 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Polynomial Approximation Optimal approximating polynomial of degree K : pa (x) =
K X
ak pk (x)
k=0
Properties of pk : different ascending degrees leading coefficient one pair wise orthogonal (inner product) Most Likely Approximating Polynomial 1200 Most Likely Polynomial Average Monday Monday 17−11−08 Monday 12−01−09
Page Hits
1000 800 600 400 200 0
6
8
10
12
14
16 Hours
18
20
22
0
2
4
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 8 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Workload Generation Unchanging Behavior Evaluate polynomial at certain points in time Random Generation Time series as point in shape space Weighting factors ak are normally distributed Multivariate Gaussian to generate weighting vector a Orthogonial Coefficients
Monomial Coefficients 0
2 1.5
−10 slope
x1
1 0.5
−20
0 −30 −0.5 −1 −5
0
5
10 x0
15
20
25
−40 0
200
400
600 average
800
1000
1200
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 9 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Benchmarking Objectives Robustness
Basic Performance
Introduction of outliers Measure over adaptation
Shifting workloads Measure peak performance
Energy and Space Efficiency
Adaptivity
Shifting workloads Measure energy / space efficiency
Different daily patterns Measure adaptability Daily Website Accesses 24 October 08 − 11 June 09
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 10 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Conclusion Conclusion Online eLearning management benchmark New generator model for query workloads More realistic benchmarking for automatic / autonomic tuning
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 11 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Conclusion Conclusion Online eLearning management benchmark New generator model for query workloads More realistic benchmarking for automatic / autonomic tuning
Future Work Tune and test Apply techniques to standard benchmarks Trends in workloads Schema evolution
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 11 / 12
Introduction
Benchmark Design
Workload Generation
Benchmarking Objectives
Conclusion
Questions?
Thank you.
Generating Shifting Workloads to Benchmark Adaptability in Relational Database Systems 12 / 12