Presentation is loading. Please wait.

Presentation is loading. Please wait.

© 1999 FORWISS FORWISS MISTRAL Performance of TPC-D Benchmark and Datawarehouses Prof. R. Bayer, Ph.D. Dr. Volker Markl Dept. of Computer Science, Technical.

Similar presentations


Presentation on theme: "© 1999 FORWISS FORWISS MISTRAL Performance of TPC-D Benchmark and Datawarehouses Prof. R. Bayer, Ph.D. Dr. Volker Markl Dept. of Computer Science, Technical."— Presentation transcript:

1 © 1999 FORWISS FORWISS MISTRAL Performance of TPC-D Benchmark and Datawarehouses Prof. R. Bayer, Ph.D. Dr. Volker Markl Dept. of Computer Science, Technical University Munich and Bavarian Research Center for Knowledgebased Systems (FORWISS)

2 © 1999 FORWISS FORWISS Test Bed for Performance Measurements l Hardware – Compaq Proliant 5000 – 4 Pentium II 200 MHz – 512 MB RAM – hard disk: 7 * 4 GB = 28 GB l Operating System – Windows NT 4.0 l RDBMS – Oracle 8 – 8kB pages l Access Methods – Tetris Algorithm for UB-Trees – Oracle IOT (clustering B*-Tree) – Oracle FTS (full table scan)

3 © 1999 FORWISS FORWISS TPC-D Schema

4 © 1999 FORWISS FORWISS Shipping Priority Query (Q3) SELECT L_ORDERKEY, SUM(L_EXTENDEDPRICE*(1-L_DISCOUNT)) AS REVENUE, O_ORDERDATE, O_SHIPPRIORITY FROM CUSTOMER, ORDER, LINEITEM WHERE C_MKTSEGMENT = 'FOOD' AND C_CUSTKEY = O_CUSTKEY AND L_ORDERKEY = O_ORDERKEY AND O_ORDERDATE DATE 1.6.98 GROUP BY L_ORDERKEY, O_ORDERDATE, O_SHIPPRIORITY ORDER BY REVENUE DESC, O_ORDERDATE

5 © 1999 FORWISS FORWISS Tetris algorithm Q3

6 © 1999 FORWISS FORWISS Response times 50% LINEITEM (Q3)

7 © 1999 FORWISS FORWISS Temporary Storage 50% LINEITEM (Q3)

8 © 1999 FORWISS FORWISS Sorting 50% of LINEITEM

9 © 1999 FORWISS FORWISS Forecasting Revenue Change Query (Q6) SUM(L_EXTENDEDPRICE*L_DISCOUNT) AS REVENUE FROM LINEITEM WHERE L_SHIPDATE >= [date] AND L_SHIPDATE <= [date] + INVERVAL 1 YEAR AND L_DISCOUNT BETWEEN [discount] -0.01 AND [discount] + 0.01 AND L_QUANTITY < [quantity]

10 © 1999 FORWISS FORWISS Forecasting Revenue Change Query (Q6)

11 © 1999 FORWISS FORWISS Performance of Q6

12 © 1999 FORWISS FORWISS Retrieving 3,3% of LINEITEM

13 © 1999 FORWISS FORWISS GFK Snowflake Schema

14 © 1999 FORWISS FORWISS TETRIS & MHC

15 © 1999 FORWISS FORWISS Performance Measurements GFK l DBMS – TransBase (covering, clustering compound B*-Trees) – UB/API on top of TransBase (UB-Tree, two ESQL Statements are optimized and processed per UB-Tree page access) – TransBase Hypercube (UB-Tree inside the DBMS Kernel) l Database – real world data warehouse from GFK – 3D Snowflake Schema »Time (3 years = 18 MP) »Segment (10500 outlets) »Product (~ 500000 items in 604 product groups) – 42 Mio fact tuples (~ 4 GB fact table size) l Computer – Sun ULTRA 1 Workstation (64 MB Main Memory)

16 © 1999 FORWISS FORWISS Indexes l MHC to encode hierarchies: – TIME_CS (5 bits) – SEGMENT_CS (24 bits) – PRODUCT_CS (29 bits) l Compound on (PRODUCT_CS, TIME_CS, SEGMENT_CS) or (TIME_CS, SEGMENT_CS, PRODUCT_CS) l UB-Tree (UB/API) on {TIME_CS, PRODUCT_CS, SEGMENT_CS}

17 © 1999 FORWISS FORWISS GFK Datawarehouse Reports  selectivity << 1%

18 © 1999 FORWISS FORWISS Compound: fixed 2MP, varying PG

19 © 1999 FORWISS FORWISS UB-Tree: fixed 2MP, varying PG

20 © 1999 FORWISS FORWISS Response Time & Result Set Size

21 © 1999 FORWISS FORWISS Clustering of UB-Trees Ø = 0.85 s / d c clustering factor

22 © 1999 FORWISS FORWISS Clustering depending on Result Set Size

23 © 1999 FORWISS FORWISS Summary UB-Tree l Excellent performance on large real DBs, > factor 10 l Very low storage requirement l 1st answer extremely fast, interactive use!! l Response time proportional to size of answer l Wide applicability: all DBs are multidimensional!! l Easy integration into DBMS, simple DDL extension l Very useful as middleware ? Patent applications


Download ppt "© 1999 FORWISS FORWISS MISTRAL Performance of TPC-D Benchmark and Datawarehouses Prof. R. Bayer, Ph.D. Dr. Volker Markl Dept. of Computer Science, Technical."

Similar presentations


Ads by Google