Kaskade: Graph Views for Efficient Graph Analytics

Automatic View Selection in Graph Databases

“…(3) Kaskade: The third algorithm is a method used by Kaskade [7]. We implemented it as follows: (i) we input the view templates with no containment to simulate its view enumeration; (ii) we enumerate the queries and evaluate the benefit of a view that contains the current query to simulate its single-view rewriting; (iii) we leverage the PROFILE [24] step to derive the size vector; and finally (iv) we use a branch-and-bound solver to select the views.…”

“…Regarding view selection in graph databases, Fan et al [9] studied the minimal and minimum containment problems but they considered the views were precomputed and static, leading to duplicate view content. Kascade [7] considered the view selection problem as an 0-1 Knapsack problem, which generated the candidates using constraint-based view enumeration, and it used a branch-and-bound solver to select the views.…”

“…However, they are not suitable for graph view selection because they do not consider the structural properties of graph queries, e.g., subgraph patterns. Kaskade [7] is a view selection method that inputs the view templates and then generates views as Cypher queries. It modeled the view selection problem as an 0-1 Knapsack problem, and used a branch-and-bound solver to select the graph views.…”

Automatic View Selection in Graph Databases

“…It is also surprising that the amount of database research literature in graph view selection is so scarce despite graph databases have become prevalent in graph data management. Particularly, Kaskade [13] inputs the view templates and then generates views as Cypher [3] queries. It modeled the view selection problem as an 0-1 Knapsack problem, and used a branch-and-bound solver to select the graph views.…”

“…The second limitation is that existing methods cannot effectively explore the possible candidate view combinations to reduce the view space and improve the view benefit. For instance, Kaskade [13] can select a single view that rewrites a given query with the highest benefit, but do not consider selecting a view set V to rewrite a query. Such a view set V could be reused to answer other contained queries, thereby saving the view space.…”