A t S + b r t T B (h i + 1) (t S + t T ) C h i (t S + t T ) + t S + b t T D (h i + n) (t S + t T )

Size: px

Start display at page:

Download "A t S + b r t T B (h i + 1) (t S + t T ) C h i (t S + t T ) + t S + b t T D (h i + n) (t S + t T )"

Raymond Williamson
5 years ago
Views:

1 Suppose we have a primary B+-tree index where the leaves contain search keys and RIDs, and the RIDs point to the records in a file that is ordered on the index search key. Assume that the blocks in the file are arranged sequentially. Which of the following is the best approximation the cost for an equality query using this index? h i = height of index t T = block transfer time t S = block seek time b r = number of blocks in file / relation b = number of blocks containing search key n = number of records containing search key A t S + b r t T B (h i + 1) (t S + t T ) C h i (t S + t T ) + t S + b t T D (h i + n) (t S + t T )

2 What is the cost for a comparison query, when using a primary B+ tree index when the search key is not unique in the data? Assume the entire relation is stored sequentially on disk. Let h i = height of index t T = block transfer time t S = block seek time b r = number of blocks in file b = number of blocks containing search key n = number of records containing search key A t S + b r t T B (h i + 1) (t S + t T ) C h i (t S + t T ) + t S + b t T D (h i + n) (t S + t T )

3 The textbook assumes that a file is always stored sequentially on disk, for purposes of demonstrating cost estimates. Which answer below would change if the file were stored nonsequentially? (Additional question for discussion: what do each of these cost estimates represent?) h i = height of index t T = block transfer time t S = block seek time b r = number of blocks in file b = number of blocks containing search key n = number of records containing search key A t S + b r t T B (h i + 1) (t T + t S ) C h i (t T + t S ) + b t T D (h i + n) (t T + t S )

4 Which of the following is the best reason as to why cost estimates (like we ve been doing) are important? A We re computer scientists, big-o-style calculations are what we do. B They allow a database system to provide for the user an estimate of how long a query will take to run. C They allow a database system to provide for the user an estimate of how much memory a query will take to run. D They help a database system administrator figure out if indexes need to be built, or if some other reorganization of the data needs to be made. E They help a database system figure out which approach to use in evaluating a query.

5 When we estimate the cost for a join, why don t we include the cost for writing out the results? Indicate all which are true. [short answer style] A It varies what we do with the results. Sometimes we write to disk, sometimes we display on the screen, sometimes we aggregate on-the-fly and just compute a total. B We don t really ultimately care about join cost; what we care about is comparing costs between joins. For all join algorithms on the same data, the cost for writing out results is the same. C Writing out the results happens in memory, so the cost is negligible. D The output from most joins in practice is small, so the cost of writing it is negligible. E It can t be cleanly estimated because it depends entirely on the results of the particular join under consideration.

6 Suppose that we wish to do a simple nested loops join on R and S, with R as the outer relation. Let: b r = blocks in R n r = tuples in R b s = blocks in s n s = tuples in s Which of the following is the best estimate of the number of blocks read? A b s + n r b r B b r + n s b s C b s + n s b r D b r + n r b s

7 Suppose that we wish to do a block nested loops join on R and S, with R as the outer relation. Let: b r = blocks in R n r = tuples in R b s = blocks in s n s = tuples in s Which of the following is the best estimate of the number of blocks read? A b s + b r b r B b r + b s b s C b s + b s b r D b r + b r b s

8 What advantages are there in using pipelining over materialized evaluation? A It produces more accurate results B It is easier to implement C It is capable of evaluating more queries than materialized evaluation can D You save the costs in writing intermediate results E You get the first tuples right away, instead of waiting for the entire query to be done before you can see anything

9 What advantages are there for demand-driven pipelining over producer-driven pipelining? A It produces more accurate results B It is easier to implement and debug C It takes advantages of multiple cores / processors

Cost-based plan selection

Cost-based plan selection Exercise 1.1 (refer to the handouts for the full exercise) σ a=1 AND b=2 AND d=3 (R) Give the best physical plan (index scan or table scan, possibly followed by a filter) and the cost of the selection.