1/24/17 3 Database Problems in Data Visualization Management Systems VIZ DATA Eugene Wu Fotis Psallidas, Zhengjie Miao, Haoci Zhang,Laura Rettig, Yifan Wu, Larry Xu, Thibault Sellam, Remco Chang, Joe Hellerstein 3 3 Database Problems in Visualization 1.Lineage 2.Consistency “Relational” 3.Query Mining 4 6 Lineage + Interactions Lineage + Interactions Lineage: high materialization cost, queries slow Lineage: high materialization cost, queries slow Interactions expressed as lineage queries ID i1, o2 backward_trace() ID ID ID ID ID i1, o3 Revenue i3, o2 Result 1 i3, o4 Profit i3, o5 Price Result 2 Product forward_trace() view_refresh() 7 1
1/24/17 Lineage + Interactions Lineage + Interactions Lineage: high materialization cost, queries slow ⨝ Revenue backward_trace( ) Interactions expressed as lineage queries Profit IDs? backward_trace() backward_trace() IDs! Revenue Revenue Profit Profit Price Price Price Price Product Product Product Product view_refresh() view_refresh() ✘ backward_trace(..) 3 Database Problems in Visualization People don’t want raw lineage ✔ query(backward_trace(..)) They want to query lineage 1.Lineage avoid materialization costs lineage query optimization 2.Consistency backward_trace() 3.Query Mining Revenue Profit Price Price Product Product 11 query 13 Perception Push-down Eyes not perfect Existing Work à Render approx viz Model human properties In General à Use models in viz system 14 2
1/24/17 No CC Serial Order Serial Order No CC Does asynchrony affect users? When? Hypotheses 1. User task matters 2. Design tricks can improve completion speed Multi-View Concurrency Control 16 17 3 Database Problems in Visualization Task matters Users self-serialize MVCC (design) has an effect 1.Lineage No Delay Delay 2.Consistency No CC 3.Query Mining Serial MVCC 10 30 50 10 30 50 T ask Completion Time 21 23 What Interfaces to Build? 2 underserved issues Many useful UI don’t exist due to high friction or org challenges SELECT * Viz systems go for FROM blackholedata coverage over task efficiency WHERE 1.2 < x AND x < 1.5 AND 40 < y AND y < 60 AND PI Precision Interfaces timestep = -50 http://ngoix.github.io/cyg-x1/index.html 3
1/24/17 SELECT * SELECT * FROM blackholedata FROM blackholedata WHERE 1.2 < x AND x < 1.5 AND WHERE 1.5 < x AND x < 1.8 AND 40 < y AND y < 60 AND 45 < y AND y < 65 AND timestep = now timestep = now http://ngoix.github.io/cyg-x1/index.html http://ngoix.github.io/cyg-x1/index.html SQL 28 29 (c) (d) (a) (b) 30 31 4
1/24/17 Sloan Digital Sky Survey Cost ~ UIComplexity + #Nodes + #Edges Simplicity Completeness 33 34 3 Database Problems in Visualization N Database Problems in Visualization 1.Lineage 1.Lineage 2.Consistency 2.Consistency 3.Query Mining 3.Query Mining 4.And more… 35 36 eugenewu.net ewu@cs.columbia.edu 5
Recommend
More recommend