Presentation is loading. Please wait.

Presentation is loading. Please wait.

Xiaodan Wang Department of Computer Science Johns Hopkins University Processing Data Intensive Queries in Scientific Database Federations.

Similar presentations


Presentation on theme: "Xiaodan Wang Department of Computer Science Johns Hopkins University Processing Data Intensive Queries in Scientific Database Federations."— Presentation transcript:

1 Xiaodan Wang Department of Computer Science Johns Hopkins University Processing Data Intensive Queries in Scientific Database Federations

2 Problem Data avalanche in scientific databases – Exponential growth in data size (Pan-STARRS) – Accumulation of data at multiple data sources (clustered and federated databases) Exploring massive, widely distributed data – Joins to find correlations across multiple databases – Queries are data intensive: large transfers over the network, and scan large portions of the data – Query throughput limits scale of exploration To improve overall query throughput but potentially sacrifice performance of individual queries

3 Processing Data Intensive Queries in Scientific Database Federations Target Application SkyQuery Federation of Astronomy Databases – Dozens of multi-terabyte databases across three Continents – Queries that perform full db scans lasting hours or days – Intermediate join results that are hundreds of MBs – Scalability concerns both in data size and number of sites Cross-match – Probabilistic spatial join across multiple databases – Join results are accumulated, shipped from site to site, and delivered to scientists

4 Processing Data Intensive Queries in Scientific Database Federations Cross-Match Workload A forward looking analysis shows that network dominates 90% of performance A quarter of the cross- match queries execute for minutes to several hours

5 Processing Data Intensive Queries in Scientific Database Federations Incorporating Network Structure

6 Processing Data Intensive Queries in Scientific Database Federations Network-Aware Join Processing Capture heterogeneity in global-scale federations – Metric to exploit high throughput paths – Decentralized, local optimizations using aggregate stats – Routing at the application layer – Two-approximate, MST-based solution with extensions that employ semi-joins and explore bushy plans – Clustering to explore trade-offs with computation cost Over a ten-fold reduction in network utilization for large joins

7 Processing Data Intensive Queries in Scientific Database Federations A Case for Batch Processing Top ten buckets accessed by 61% of queries and reuse occur close temporally 2% of buckets capture more than half of the workload and should be cached

8 Processing Data Intensive Queries in Scientific Database Federations LifeRaft: Data-Driven Batch Proc. Eliminate redundant I/O to improve query throughput Batch queries with that exhibit data sharing – Pre-process queries to identify data sharing – Co-schedule queries that access the same data – Access contentious data first to maximize sharing – Improves performance by two-fold

9 Processing Data Intensive Queries in Scientific Database Federations Discussion Cache replacement for LifeRaft – Benefits contentions data regions that experience reuse (Cache hit for LifeRaft is 40% compared with 7% for arrival order processing) – Evaluate strategies that exploit I/O behavior of batch workloads (segmented strategy) Buffering and workload overflow – Large intermediate join results – Migrate pairs of workload and bucket Better support for interactive queries – Short and selective queries that focus on small region – Indefinite queuing times in presence of batch workloads

10 Processing Data Intensive Queries in Scientific Database Federations Discussion (cont.) Batch processing in a distributed environment – Network-aware scheduling does not consider computation cost – Batch processing for a single system environment Federating LifeRaft – Coordinate exec. of query that join multiple DBs – Batch proc. requires databases to buffer results – Maximize overall batch size while alleviating memory used for buffering and network cost

11 Processing Data Intensive Queries in Scientific Database Federations Exploring Alt. Join Schedules

12 Processing Data Intensive Queries in Scientific Database Federations Discussion (cont.) Explore both join schedules and opportunities for batching simultaneously – Bushy and semi-join plans increase computation while clustering decrease computation – Skew in join workload (ie. sites close to end user) – Quantify trade-offs with computation cost (ie. number of buckets in batch processing) Users submit cross-match queries in batches Applying LifeRaft to other data-intensive, temporal- spatial data such as Turbulence database

13 Processing Data Intensive Queries in Scientific Database Federations Supplementary Slides

14 Processing Data Intensive Queries in Scientific Database Federations Cross-Match Queries Join by increasing cardinality (count *) – Minimal I/O – Fewer bytes on the network Query Mediator Probe Query Result Count: 30Count: 100Count: 800

15 Processing Data Intensive Queries in Scientific Database Federations Spanning Tree Approximation (STA) B C A D E F G H

16 Processing Data Intensive Queries in Scientific Database Federations STA: Find MST B C A D E F G H

17 Processing Data Intensive Queries in Scientific Database Federations STA: Join Using Paths on the MST B C A D E F G H 1 2 3 5 4 6 7 9 8 10 12 11 13

18 Processing Data Intensive Queries in Scientific Database Federations Filter and refine Partition data into buckets

19 Processing Data Intensive Queries in Scientific Database Federations Scheduling Behavior Q i – Q i1, Q i2, Q i3 B1B1 B2B2 B3B3 B4B4 B5B5 B6B6 B7B7 B8B8 QiQi QjQj QkQk Sub-divide queries by bucket: Q j – Q j3, Q j4, Q j5, Q j6, Q j7, Q j8 Assumptions: - Inter-query time of 1 sec - I/O for each bucket of 1 sec - Cache size of 2 - Join cost is negligible Q j – Q j5, Q j6, Q j7, Q j8 QkQk

20 Processing Data Intensive Queries in Scientific Database Federations Arrival order with no sharing Qi1Qi1 B1B1 Q i Arr Qi2Qi2 B2B2 Qi3Qi3 B3B3 Qj1Qj1 B1B1 Q j ArrQ k Arr Qj3Qj3 B3B3 Q i End Qj4Qj4 B4B4 Qj6Qj6 B6B6 Qj7Qj7 B7B7 Qj8Qj8 B8B8 Q j End Qk1Qk1 B1B1 Qk4Qk4 B4B4 Qk8Qk8 B8B8 Q k End Q i – 3 sec Completion Times: Q j – 8 secQ k – 13 secAvg – 8 sec B1B1 B2B2 B3B3 B4B4 B5B5 B6B6 B7B7 B8B8 QiQi QjQj QkQk QkQk … Tp –.2 qry/sec

21 Processing Data Intensive Queries in Scientific Database Federations Age based scheduling (bias 1) Qi1Qi1 B1B1 Q i Arr Qi2Qi2 B2B2 Qi5Qi5 B5B5 Qi3Qj3Qi3Qj3 B3B3 Q j ArrQ k ArrQ i End Q j End Q k End Qj1Qk1Qj1Qk1 B1B1 Qj4Qk4Qj4Qk4 B4B4 Qj6Qk6Qj6Qk6 B6B6 Q i – 3 sec Completion Times: Q j – 7 secQ k – 7 secAvg – 5.6 secTp –.33 qry/sec B1B1 B2B2 B3B3 B4B4 B5B5 B6B6 B7B7 B8B8 QiQi QjQj QkQk QkQk Qj8Qk8Qj8Qk8 B8B8 Qj7Qk7Qj7Qk7 B7B7

22 Processing Data Intensive Queries in Scientific Database Federations Contention based scheduling (bias 0) Qi1Qi1 B1B1 Q i Arr Qi2Qi2 B2B2 Qi3Qj3Qi3Qj3 B3B3 Q j ArrQ k Arr Q i End Q j End Qk5Qk5 B5B5 Q k End Q j1 Q k1 Q j4 Q k4 B 1 B 4 Qj6Qk6Qj6Qk6 B6B6 Qj7Qk7Qj7Qk7 B7B7 Q i – 7 sec Completion Times: Q j – 5 secQ k – 6 secAvg – 6 secTp –.38 qry/sec B1B1 B2B2 B3B3 B4B4 B5B5 B6B6 B7B7 B8B8 QiQi QjQj QkQk QkQk Qj8Qk8Qj8Qk8 B8B8 (5.6) (.33)

23 Processing Data Intensive Queries in Scientific Database Federations Parameter tuning using trade-off curves

24 Processing Data Intensive Queries in Scientific Database Federations Tuning the age bias Throughput performance gap grows while response time gap is insensitive to saturation Increasing age bias is more attractive at low saturation


Download ppt "Xiaodan Wang Department of Computer Science Johns Hopkins University Processing Data Intensive Queries in Scientific Database Federations."

Similar presentations


Ads by Google