Paper Review DBToaster Higher order Delta Processing for Dynamic, Frequently Fresh Views∗



Viewlet transforms, a recursive finite differencing techniques applied to queries. The viewlet transform materializes a query and a set of its higher-order deltas as views.

Incremental View Maintenance — higher-order IVM.


The system supports tens of thousands of complete view refreshes a second for a wide range of queries.


Related theory work — discrete wavelets and numerical differentiation methods. “Joins are only needed in the presence of inequality joins and nested aggregates in view definitions”.

DBToaster’s comment on PL work in this area: “Automatic incrementalization is by no means a solved challenge, esp. when considering general recursive and unbounded iteration"

Mapping from tuples to multiplicities — key abstraction. Use multiplicities to represent aggregate query results.

Delta rules for operators — I need to encode that in Cozy.

Binding pattens — the subquery can only be computed if a value for the input variable is given.

Related work: trigger in view updates.

Optimizing viewlets:

  1. Materialization decisions
    1. Duplicate View Elimination
    2. Query Decomposition
    3. Polynomial Expansion nad Factorization
    4. Input Variables
    5. Details of Nested Aggregates
    6. Cost model

Relationship between sub-query, graphical models and polynomials.

Practical Value

What you can learn from this to make your research better?

Details and Problems From the presenters’ point of view, what questions might audience ask?

Why is it finite? Any proof in paper about that?

Parallelization — what could we do in Cozy? What about stream processing?

What if Cozy can accumulate deltas and perform update in slump sum?

How it deals with numerical distribution laws?