Title: Storage, Indexing and Joins Slides taken from Database Management Systems, R. Ramakrishnan
1Storage, Indexing and Joins Slides taken
fromDatabase Management Systems, R.
Ramakrishnan
2Disks and Files
- DBMS stores information on (hard) disks.
- This has major implications for DBMS design!
- READ transfer data from disk to main memory
(RAM). - WRITE transfer data from RAM to disk.
- Both are high-cost operations, relative to
in-memory operations, so must be planned
carefully!
3Why Not Store Everything in Main Memory?
- Costs too much. 1000 will buy you either 0.5GB
of RAM or 50GB of disk today. - Main memory is volatile. We want data to be
saved between runs. (Obviously!) - Typical storage hierarchy
- Main memory (RAM) for currently used data.
- Disk for the main database (secondary storage).
- Tapes for archiving older versions of the data
(tertiary storage).
4Disks
- Secondary storage device of choice.
- Main advantage over tapes random access vs.
sequential. - Data is stored and retrieved in units called disk
blocks or pages. - Unlike RAM, time to retrieve a disk page varies
depending upon location on disk. - Therefore, relative placement of pages on disk
has major impact on DBMS performance!
5Components of a Disk
Spindle
Disk head
- The platters spin (say, 90rps).
- The arm assembly is moved in or out to position
a head on a desired track. Tracks under heads
make a cylinder (imaginary!).
Sector
Platters
- Only one head reads/writes at any one time.
- Block size is a multiple of sector size (which
is fixed).
6Accessing a Disk Page
- Time to access (read/write) a disk block
- seek time (moving arms to position disk head on
track) - rotational delay (waiting for block to rotate
under head) - transfer time (actually moving data to/from disk
surface) - Seek time and rotational delay dominate.
- Seek time varies from about 1 to 20msec
- Rotational delay varies from 0 to 10msec
- Transfer rate is about 1msec per 4KB page
- Key to lower I/O cost reduce seek/rotation
delays! Hardware vs. software solutions?
7Arranging Pages on Disk
- Next block concept
- blocks on same track, followed by
- blocks on same cylinder, followed by
- blocks on adjacent cylinder
- Blocks in a file should be arranged sequentially
on disk (by next), to minimize seek and
rotational delay. - For a sequential scan, pre-fetching several pages
at a time is a big win!
8Disk Space Management
- Lowest layer of DBMS software manages space on
disk. - Higher levels call upon this layer to
- allocate/de-allocate a page
- read/write a page
- Request for a sequence of pages must be satisfied
by allocating the pages sequentially on disk!
Higher levels dont need to know how this is
done, or how free space is managed.
9Buffer Management in a DBMS
Page Requests from Higher Levels
BUFFER POOL
disk page
free frame
MAIN MEMORY
DISK
choice of frame dictated by replacement policy
- Data must be in RAM for DBMS to operate on it!
- Table of ltframe, pageidgt pairs is maintained.
10When a Page is Requested ...
- If requested page is not in pool
- Choose a frame for replacement
- If frame is dirty, write it to disk
- Read requested page into chosen frame
- Pin the page and return its address.
- If requests can be predicted (e.g., sequential
scans) - pages can be pre-fetched several pages at a
time!
11More on Buffer Management
- Requestor of page must unpin it, and indicate
whether page has been modified - dirty bit is used for this.
- Page in pool may be requested many times,
- a pin count is used. A page is a candidate for
replacement iff pin count 0. - CC recovery may entail additional I/O when a
frame is chosen for replacement.
12Buffer Replacement Policy
- Frame is chosen for replacement by a replacement
policy - Least-recently-used (LRU), Clock, MRU etc.
- Policy can have big impact on of I/Os depends
on the access pattern. - Sequential flooding Nasty situation caused by
LRU repeated sequential scans. - buffer frames lt pages in file means each page
request causes an I/O. MRU much better in this
situation (but not in all situations, of course).
13DBMS vs. OS File System
- OS does disk space buffer management why
not let OS manage these tasks? - Differences in OS support portability issues
- Some limitations, e.g., files cant span disks.
- Buffer management in DBMS requires ability to
- pin a page in buffer pool, force a page to disk
(important for implementing CC recovery), - adjust replacement policy, and pre-fetch pages
based on access patterns in typical DB operations.
14Files of Records
- Page or block is OK when doing I/O, but higher
levels of DBMS operate on records, and files of
records. - FILE A collection of pages, each containing a
collection of records. Must support - insert/delete/modify record
- read a particular record (specified using record
id) - scan all records (possibly with some conditions
on the records to be retrieved)
15Record Formats Fixed Length
F1
F2
F3
F4
L1
L2
L3
L4
Base address (B)
Address BL1L2
- Information about field types same for all
records in a file stored in system catalogs. - Finding ith field requires scan of record.
16Record Formats Variable Length
- Two alternative formats ( fields is fixed)
F1 F2 F3
F4
Fields Delimited by Special Symbols
Field Count
F1 F2 F3 F4
Array of Field Offsets
- Second offers direct access to ith field,
efficient storage - of nulls (special dont know value) small
directory overhead.
17Page Formats Fixed Length Records
Slot 1
Slot 1
Slot 2
Slot 2
Free Space
. . .
. . .
Slot N
Slot N
Slot M
N
M
1
. . .
1
1
0
M ... 3 2 1
number of records
number of slots
PACKED
UNPACKED, BITMAP
- Record id ltpage id, slot gt. In first
alternative, moving records for free space
management changes rid may not be acceptable.
18Page Formats Variable Length Records
Rid (i,N)
Page i
Rid (i,2)
Rid (i,1)
N
Pointer to start of free space
20
16
24
N . . . 2 1
slots
SLOT DIRECTORY
- Can move records on page without changing rid
so, attractive for fixed-length records too.
19Alternative File Organizations
- Many alternatives exist, each ideal for some
situation, and not so good in others - Heap files Suitable when typical access is a
file scan retrieving all records. - Sorted Files Best if records must be retrieved
in some order, or only a range of records is
needed. - Hashed Files Good for equality selections.
- File is a collection of buckets. Bucket primary
page plus zero or more overflow pages. - Hashing function h h(r) bucket in which
record r belongs. h looks at only some of the
fields of r, called the search fields.
20Unordered (Heap) Files
- Simplest file structure contains records in no
particular order. - As file grows and shrinks, disk pages are
allocated and de-allocated. - To support record level operations, we must
- keep track of the pages in a file
- keep track of free space on pages
- keep track of the records on a page
- There are many alternatives for keeping track of
this.
21Heap File Implemented as a List
Data Page
Data Page
Data Page
Full Pages
Header Page
Data Page
Data Page
Data Page
Pages with Free Space
- The header page id and Heap file name must be
stored someplace. - Each page contains 2 pointers plus data.
22Heap File Using a Page Directory
- The entry for a page can include the number of
free bytes on the page. - The directory is a collection of pages linked
list implementation is just one alternative. - Much smaller than linked list of all heap file
pages!
23Analysis of file organizations
- We ignore CPU costs for simplicity, and use the
following parameters in our cost model - B The number of data pages
- R Number of records per page
- D (Average) time to read or write disk page
- Measuring number of page I/Os ignores gains of
pre-fetching blocks of pages thus, even I/O cost
is only approximated. - Average-case analysis based on several
simplistic assumptions.
- Good enough to show the overall trends!
24Assumptions in Our Analysis
- Single record insert and delete.
- Heap Files
- Equality selection on key exactly one match.
- Insert always at end of file.
- Sorted Files
- Files compacted after deletions.
- Selections on sort field(s).
- Hashed Files
- No overflow buckets, 80 page occupancy.
25Cost of Operations
- Several assumptions underlie these (rough)
estimates!
26Indexes
- A Heap file allows us to retrieve records
- by specifying the rid, or
- by scanning all records sequentially
- Sometimes, we want to retrieve records by
specifying the values in one or more fields,
e.g., - Find all students in the CS department
- Find all students with a gpa gt 3
- Indexes are file structures that enable us to
answer such value-based queries efficiently. - This will be topic of our next lecture!
27 28Indexes
- An index on a file speeds up selections on the
search key fields for the index. - Any subset of the fields of a relation can be the
search key for an index on the relation. - Search key is not the same as key (minimal set of
fields that uniquely identify a record in a
relation). - An index contains a collection of data entries,
and supports efficient retrieval of all data
entries k with a given key value k.
29Alternatives for Data Entry k in Index
- Three alternatives
- Data record with key value k
- ltk, rid of data record with search key value kgt
- ltk, list of rids of data records with search key
kgt - Choice of alternative for data entries is
orthogonal to the indexing technique used to
locate data entries with a given key value k. - Examples of indexing techniques B trees,
hash-based structures - Typically, index contains auxiliary information
that directs searches to the desired data entries
30Index Classification
- Primary vs. secondary If search key contains
primary key, then called primary index. - Unique index Search key contains a candidate
key. - Clustered vs. unclustered If order of data
records is the same as, or close to, order of
data entries, then called clustered index. - Alternative 1 implies clustered, but not
vice-versa. - A file can be clustered on at most one search
key. - Cost of retrieving data records through index
varies greatly based on whether index is
clustered or not!
31Clustered vs. Unclustered Index
- Suppose that Alternative (2) is used for data
entries, and that the data records are stored in
a Heap file. - To build clustered index, first sort the Heap
file (with some free space on each page for
future inserts). - Overflow pages may be needed for inserts. (Thus,
order of data recs is close to, but not
identical to, the sort order.)
Index entries
UNCLUSTERED
CLUSTERED
direct search for
data entries
Data entries
Data entries
(Index File)
(Data file)
Data Records
Data Records
32Index Classification (Cont.)
- Dense vs. Sparse If there is at least one data
entry per search key value (in some data
record), then dense. - Alternative 1 always leads to dense index.
- Every sparse index is clustered!
- Sparse indexes are smaller however, some useful
optimizations are based on dense indexes.
Ashby, 25, 3000
22
Basu, 33, 4003
25
Bristow, 30, 2007
30
Ashby
33
Cass, 50, 5004
Cass
Smith
Daniels, 22, 6003
40
Jones, 40, 6003
44
44
Smith, 44, 3000
50
Tracy, 44, 5004
Sparse Index
Dense Index
on
on
Data File
Name
Age
33Range Searches
- Find all students with gpa gt 3.0
- If data is in sorted file, do binary search to
find first such student, then scan to find
others. - Cost of binary search can be quite high.
- Simple idea Create an index file.
Index File
kN
k2
k1
Data File
Page N
Page 3
Page 1
Page 2
- Can do binary search on (smaller) index file!
34Comments on ISAM
Data Pages
Index Pages
- File creation Leaf (data) pages allocated
sequentially, sorted by search key
then index pages allocated, then space for
overflow pages. - Index entries ltsearch key value, page idgt
they direct search for data entries, which
are in leaf pages. - Search Start at root use key comparisons to go
to leaf. Cost log F N F entries/index
pg, N leaf pgs - Insert Find leaf data entry belongs to, and put
it there. - Delete Find and remove from leaf if empty
overflow page, de-allocate.
Overflow pages
- Static tree structure inserts/deletes affect
only leaf pages.
35B Tree The Most Widely Used Index
- Insert/delete at log F N cost keep tree
height-balanced. (F fanout, N leaf pages) - Minimum 50 occupancy (except for root). Each
node contains d lt m lt 2d entries. The
parameter d is called the order of the tree. - Supports equality and range-searches efficiently.
36Example B Tree
- Search begins at root, and key comparisons direct
it to a leaf. - Search for 5, 15, all data entries gt 24 ...
Root
17
24
30
13
39
3
5
19
20
22
24
27
38
2
7
14
16
29
33
34
- Based on the search for 15, we know it is not
in the tree!
37Inserting a Data Entry into a B Tree
- Find correct leaf L.
- Put data entry onto L.
- If L has enough space, done!
- Else, must split L (into L and a new node L2)
- Redistribute entries evenly, copy up middle key.
- Insert index entry pointing to L2 into parent of
L. - This can happen recursively
- To split index node, redistribute entries evenly,
but push up middle key. (Contrast with leaf
splits.) - Splits grow tree root split increases height.
- Tree growth gets wider or one level taller at
top.
38Inserting 8 into Example B Tree
Entry to be inserted in parent node.
- Observe how minimum occupancy is guaranteed in
both leaf and index pg splits. - Note difference between copy-up and push-up be
sure you understand the reasons for this.
(Note that 5 is
s copied up and
5
continues to appear in the leaf.)
3
5
2
7
8
appears once in the index. Contrast
39Example B Tree After Inserting 8
Root
17
24
30
13
5
2
3
39
19
20
22
24
27
38
7
5
8
14
16
29
33
34
- Notice that root was split, leading to increase
in height.
- In this example, we can avoid split by
re-distributing entries however,
this is usually not done in practice.
40Deleting a Data Entry from a B Tree
- Start at root, find leaf L where entry belongs.
- Remove the entry.
- If L is at least half-full, done!
- If L has only d-1 entries,
- Try to re-distribute, borrowing from sibling
(adjacent node with same parent as L). - If re-distribution fails, merge L and sibling.
- If merge occurred, must delete entry (pointing to
L or sibling) from parent of L. - Merge could propagate to root, decreasing height.
41Example Tree After (Inserting 8, Then) Deleting
19 and 20 ...
Root
17
27
30
13
5
2
3
39
38
7
5
8
22
24
27
29
14
16
33
34
- Deleting 19 is easy.
- Deleting 20 is done with re-distribution. Notice
how middle key is copied up.
42 ... And Then Deleting 24
- Must merge.
- Observe toss of index entry (on right), and
pull down of index entry (below).
30
39
22
27
38
29
33
34
Root
13
5
30
17
3
39
2
7
22
38
5
8
27
33
34
14
16
29
43Summary (cont.)
- Data entries can be actual data records, ltkey,
ridgt pairs, or ltkey, rid-listgt pairs. - Choice orthogonal to indexing technique used to
locate data entries with a given key value. - Can have several indexes on a given file of data
records, each with a different search key. - Indexes can be classified as clustered vs.
unclustered, primary vs. secondary, and dense vs.
sparse. Differences have important consequences
for utility/performance.
44Summary (cont.)
- Tree-structured indexes are ideal for
range-searches, also good for equality searches. - B tree is a dynamic structure.
- Inserts/deletes leave tree height-balanced log F
N cost. - High fanout (F) means depth rarely more than 3 or
4. - Almost always better than maintaining a sorted
file.
45Implementation of Relational Operations
46Relational Operations
- We will consider how to implement
- Selection ( ) Selects a subset of rows
from relation. - Projection ( ) Deletes unwanted columns
from relation. - Join ( ) Allows us to combine two
relations. - Set-difference ( ) Tuples in reln. 1, but
not in reln. 2. - Union ( ) Tuples in reln. 1 and in reln. 2.
- Aggregation (SUM, MIN, etc.) and GROUP BY
47Schema for Examples
Sailors (sid integer, sname string, rating
integer, age real) Reserves (sid integer, bid
integer, day dates, rname string)
- Reserves
- Each tuple is 40 bytes long, 100 tuples per
page, 1000 pages. - Sailors
- Each tuple is 50 bytes long, 80 tuples per page,
500 pages.
48Equality Joins With One Join Column
SELECT FROM Reserves R1, Sailors S1 WHERE
R1.sidS1.sid
- In algebra R S. Common! Must be
carefully optimized. R S is large so, R
S followed by a selection is inefficient. - Assume M pages in R, pR tuples per page, N pages
in S, pS tuples per page. - In our examples, R is Reserves and S is Sailors.
- We will consider more complex join conditions
later. - Cost metric of I/Os. We will ignore output
costs.
49Simple Nested Loops Join
foreach tuple r in R do foreach tuple s in S
do if ri sj then add ltr, sgt to result
- For each tuple in the outer relation R, we scan
the entire inner relation S. - Cost M (pR M) N 1000 1001000500
I/Os (a lot!) - Page-oriented Nested Loops join For each page
of R, get each page of S, and write out matching
pairs of tuples ltr, sgt, where r is in R-page
and S is in S-page. - Cost M MN 1000 1000500 501,000
- If smaller relation (S) is outer, cost 500
5001000 500,500
50Block Nested Loops Join
- Use one page as an input buffer for scanning the
inner S, one page as the output buffer, and use
all remaining pages to hold block of outer R. - For each matching tuple r in R-block, s in
S-page, add ltr, sgt to result. Then read
next R-block, scan S, etc.
R S
Join Result
Hash table for block of R (k lt B-1 pages)
. . .
. . .
Input buffer for S
Output buffer
51Examples of Block Nested Loops
- Cost Scan of outer outer blocks scan of
inner - outer blocks
- With Reserves (R) as outer, and blocksize100
- Cost of scanning R is 1000 I/Os a total of 10
blocks. - Per block of R, we scan Sailors (S) 10500
I/Os. - TOTAL 6,000 I/Os
- With 100-page block of Sailors as outer
- Cost of scanning S is 500 I/Os a total of 5
blocks. - Per block of S, we scan Reserves 51000 I/Os.
- TOTAL 5,500 I/Os
52Index Nested Loops Join
foreach tuple r in R do foreach tuple s in S
where ri sj do add ltr, sgt to result
- If there is an index on the join column of one
relation (say S), can make it the inner and
exploit the index. - Cost M ( (MpR) cost of finding matching S
tuples) - For each R tuple, cost of probing S index is
about 1.2 for hash index, 2-4 for B tree. Cost
of then finding S tuples depends on clustering. - Clustered index 1 I/O (typical), unclustered
up to 1 I/O per matching S tuple.
53Examples of Index Nested Loops
- Hash-index on sid of Sailors (as inner)
- Scan Reserves 1000 page I/Os, 1001000 tuples.
- For each Reserves tuple 1.2 I/Os to get data
entry in index, plus 1 I/O to get (the exactly
one) matching Sailors tuple. Total
221,000 I/Os. - Hash-index on sid of Reserves (as inner)
- Scan Sailors 500 page I/Os, 80500 tuples.
- For each Sailors tuple 1.2 I/Os to find index
page with data entries, plus cost of retrieving
matching Reserves tuples. Assuming uniform
distribution, 2.5 reservations per sailor. Cost
of retrieving them is 1 or 2.5 I/Os depending on
whether the index is clustered. Total 500
40,000(1.22.5) 148,500
54Sort-Merge Join (R S)
ij
- Sort R and S on the join column, then scan them
to do a merge (on join column), and output
result tuples. - R is scanned once each S group is scanned once
per matching R tuple. (Multiple scans of an S
group are likely to find needed pages in buffer.)
55Example of Sort-Merge Join
- Cost M log M N log N (MN)
- The cost of scanning, MN, could be MN (very
unlikely!) - With 35, 100 or 300 buffer pages, both Reserves
and Sailors can be sorted in 2 passes total join
cost 7500.
(BNL cost 2500 to 15000 I/Os)
56Hash-Join
- Partition both relations using hash fn h R
tuples in partition i will only match S tuples in
partition i.
- Read in a partition of R, hash it using h2 (ltgt
h!). Scan matching partition of S, search for
matches.
57Cost of Hash-Join
- In partitioning phase, readwrite both relns
2(MN). In matching phase, read both relns MN
I/Os. - In our running example, this is a total of 4500
I/Os. - Sort-Merge Join vs. Hash Join
- Given a minimum amount of memory (what is this,
for each?) both have a cost of 3(MN) I/Os. Hash
Join superior on this count if relation sizes
differ greatly. Also, Hash Join shown to be
highly parallelizable. - Sort-Merge less sensitive to data skew result is
sorted.
58General Join Conditions
- Equalities over several attributes (e.g.,
R.sidS.sid AND R.rnameS.sname) - For Index NL, build index on ltsid, snamegt (if S
is inner) or use existing indexes on sid or
sname. - For Sort-Merge and Hash Join, sort/partition on
combination of the two join columns. - Inequality conditions (e.g., R.rname lt S.sname)
- For Index NL, need (clustered!) B tree index.
- Range probes on inner matches likely to be
much higher than for equality joins. - Hash Join, Sort Merge Join not applicable.
- Block NL quite likely to be the best join method
here.
59Simple Selections
SELECT FROM Reserves R WHERE R.rname lt
C
- Of the form
- Size of result approximated as size of R
reduction factor. - With no index, unsorted Must essentially scan
the whole relation cost is M (pages in R). - With an index on selection attribute Use index
to find qualifying data entries, then retrieve
corresponding data records. (Hash index useful
only for equality selections.)
60Using an Index for Selections
- Cost depends on qualifying tuples, and
clustering. - Cost of finding qualifying data entries
(typically small) plus cost of retrieving records
(could be large w/o clustering). - In example, assuming uniform distribution of
names, about 10 of tuples qualify (100 pages,
10000 tuples). With a clustered index, cost is
little more than 100 I/Os if unclustered, up to
10000 I/Os! - Important refinement for unclustered indexes
- 1. Find qualifying data entries.
- 2. Sort the rids of the data records to be
retrieved. - 3. Fetch rids in order. This ensures that each
data page is looked at just once (though of
such pages likely to be higher than with
clustering).
61General Selection Conditions
- (daylt8/9/94 AND rnamePaul) OR bid5 OR sid3
- Such selection conditions are first converted to
conjunctive normal form (CNF)
(daylt8/9/94 OR bid5 OR
sid3 ) AND (rnamePaul OR bid5 OR
sid3) - We only discuss the case with no ORs (a
conjunction of terms of the form attr op value). - An index matches (a conjunction of) terms that
involve only attributes in a prefix of the search
key. - Index on lta, b, cgt matches a5 AND b 3, but not
b3.
62Two Approaches to General Selections
- First approach Find the most selective access
path, retrieve tuples using it, and apply any
remaining terms that dont match the index - Most selective access path An index or file scan
that we estimate will require the fewest page
I/Os. - Terms that match this index reduce the number of
tuples retrieved other terms are used to discard
some retrieved tuples, but do not affect number
of tuples/pages fetched. - Consider daylt8/9/94 AND bid5 AND sid3. A B
tree index on day can be used then, bid5 and
sid3 must be checked for each retrieved tuple.
Similarly, a hash index on ltbid, sidgt could be
used daylt8/9/94 must then be checked.
63Intersection of Rids
- Second approach (if we have 2 or more matching
indexes that use Alternatives (2) or (3) for data
entries) - Get sets of rids of data records using each
matching index. - Then intersect these sets of rids (well discuss
intersection soon!) - Retrieve the records and apply any remaining
terms. - Consider daylt8/9/94 AND bid5 AND sid3. If we
have a B tree index on day and an index on sid,
both using Alternative (2), we can retrieve rids
of records satisfying daylt8/9/94 using the first,
rids of recs satisfying sid3 using the second,
intersect, retrieve records and check bid5.
64The Projection Operation
SELECT DISTINCT R.sid,
R.bid FROM Reserves R
- An approach based on sorting
- Eliminate unwanted fields in first pass of
external sort. Thus, runs of about 2B pages are
produced, but tuples in runs are smaller than
input tuples. (Size ratio depends on and size
of fields that are dropped.) - Modify merging passes to eliminate duplicates.
Thus, number of result tuples smaller than input.
(Difference depends on of duplicates.) - Cost In first pass, read original relation
(size M), write out same number of smaller
tuples. In merging passes, fewer tuples written
out in each pass. Using Reserves example, 1000
input pages reduced to 250 in first pass if size
ratio is 0.25
65Discussion of Projection
- Sort-based approach is standard it is often
useful that the result is sorted. - There is also an approach based on hashing (see
book). - If an index on the relation contains all wanted
attributes in its search key, can do index-only
scan. - Apply projection techniques to data entries (much
smaller!) - If an ordered (i.e., tree) index contains all
wanted attributes as prefix of search key, can do
even better - Retrieve data entries in order (index-only scan),
discard unwanted fields, compare adjacent tuples
to check for duplicates.
66Set Operations
- Intersection and cross-product special cases of
join. - Union (Distinct) and Except similar well do
union. - Sorting based approach to union
- Sort both relations (on combination of all
attributes). - Scan sorted relations and merge them.
- Hash based approach to union
- Partition R and S using hash function h.
- For each S-partition, build in-memory hash table
(using h2), scan corresponding R-partition and
add tuples to table while discarding duplicates.
67Aggregate Operations (AVG, MIN, etc.)
- Without grouping
- In general, requires scanning the relation.
- Given index whose search key includes all
attributes in the SELECT or WHERE clauses, can do
index-only scan. - With grouping
- Sort on group-by attributes, then scan relation
and compute aggregate for each group. (Can
improve upon this by combining sorting and
aggregate computation.) - Similar approach based on hashing on group-by
attributes. - Given tree index whose search key includes all
attributes in SELECT, WHERE and GROUP BY clauses,
can do index-only scan if group-by attributes
form prefix of search key, can retrieve data
entries/tuples in group-by order.
68Summary
- A virtue of relational DBMSs queries are
composed of a few basic operators the
implementation of these operators can be
carefully tuned (and it is important to do
this!). - Many alternative implementation techniques for
each operator no universally superior technique
for most operators. - Must consider available alternatives for each
operation in a query and choose best one based on
system statistics, etc. This is part of the
broader task of optimizing a query composed of
several ops.