The MV3R-Tree: A Spatio-

Temporal Access Method for

Timestamp and Interval Queries

Yufei Tao and Dimitris Papadias

Hong Kong University of Science and
Technology

Present by
Guangyue Jia

1
 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

2
 Motivation
• Spatio-tempral queries are common.
– Data: (tid, sid, x1, y1, x2, y2, t1, t2)
– Query: space restriction + time restriction

• Timestamp (or timeslice) query and

interval query.

• Process both timestamp and interval query

efficiently.
3
 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

4
 Related works
• MVB-tree:
– Multi-version B-tree

• HR-tree
– Historical R-tree

• 3D R-tree
– 3 Dimension R-tree
5
 MVB-tree
• Each entry has the form: <key, tstart, tend, point>
– Insertions and deletions only happen at the current time.
• An entry called alive at a timestamp t if tstart <=t<tend
• Multiple roots and each root has a jurisdiction interval:
– Minimum bounding lifespan of all the entries in the root.
• Either none or b · pversion entries are alive for each
timestamp t and each node except the roots.
– Ensures that entries alive at the same timestamps are mostly
grouped together.

pversion =1/3
b=6

6
Example of MVB-tree
 Overflow
• Insertion is carried out as B-tree except
overflows.
• Weak version overflow: block overflow.
And it cause version split.
• number of live entries in a new node must
be in the range [b·psvu, b·psvo].
• Strong version overflow: number of live
entries exceeds b·psvo.
• Strong version overflow cause key split.
7
 Version split
Description:
1, When weak version overflow.
2, All live entries are copied to a
new node.
3, tstart modified to the current
time.
4, tend of the live entries is
changed from * to the current
time.
Example of MVB-tree 5, create data redundancy.

Example:
1, insert <28, 4, *> at time 4.
2, A weak version overflow.
3, create new node D and copy
the live entries to it.
4, all tstart are set to be 4
8
Example of block overflow and version split 5, A ”dies”, all * are replaced by 4.
 Underflow
• Deletion is carried out as B-tree except
underflows.
• Weak version underflow: number of live
entries lower than b·pversion
• Strong version underflow: number of live
entries becomes lower than b·psvu.
• Underflow: copy sibling node using only its
live entries.
9
 Mergence and key split
pversion =psvu=1/3
Root A B C
psvo=5/6
<5, 1, *> <43, 1, *> <72, 1, *>
<8, 1, *> <48, 1, *> <78, 1, *> 1, delete entry <48, 1, *>
<5, 1, *, A>
<13, 1, *> <52, 1, 2> <83, 1, *> at timestamp 4.
<43, 1, *, B>
<25, 1, 3> <59, 1, 3> <95, 1, 3>
<72, 1, *, C> 2, B weak version
<27, 1, 3> <27, 1, 3> <99, 1, *>
<39, 1, 3> <68, 1, 3> <102, 1, *> underflow since only
entry <43, 1, *> is alive.
3, copy live entries from
sibling node C to C´.
3, insert <43, 4, *> into
C´cause strong version
overflow.
4, key split and node D
and E are created.

10
 Mergence and key split
pversion =psvu=1/3
Root A B C
psvo=5/6
<5, 1, *> <43, 1, *> <72, 1, *>
<8, 1, *> <48, 1, 4> <78, 1, *> 1, delete entry <48, 1, *>
<5, 1, *, A>
<13, 1, *> <52, 1, 2> <83, 1, *> at timestamp 4.
<43, 1, *, B>
<25, 1, 3> <59, 1, 3> <95, 1, 3>
<72, 1, *, C> 2, B weak version
<27, 1, 3> <27, 1, 3> <99, 1, *>
<39, 1, 3> <68, 1, 3> <102, 1, *> underflow since only
entry <43, 1, *> is alive.
3, copy live entries from
sibling node C to C´.
3, insert <43, 4, *> into
C´cause strong version
overflow.
4, key split and node D
and E are created.

11
 Mergence and key split
pversion =psvu=1/3
Root A B C
psvo=5/6
<5, 1, *> <43, 1, *> <72, 1, *>
<5, 1, *, A> <8, 1, *> <48, 1, 4> <78, 1, *> 1, delete entry <48, 1, *>
<43, 1, *, B> <13, 1, *> <52, 1, 2> <83, 1, *> at timestamp 4.
<72, 1, *, C> <25, 1, 3> <59, 1, 3> <95, 1, 3>
<72, 1, *, C´> <27, 1, 3> <27, 1, 3> <99, 1, *> 2, B weak version
<39, 1, 3> <68, 1, 3> <102, 1, *> underflow since only
entry <43, 1, *> is alive.
3, copy live entries from
sibling node C to C´.
C´
3, insert <43, 4, *> into
<72, 1, *> C´cause strong version
<78, 1, *> overflow.
<83, 1, *>
<99, 1, *> 4, key split and node D
<102, 1, *> and E are created.

12
 Mergence and key split
pversion =psvu=1/3
Root A B C
psvo=5/6
<5, 1, *> <43, 1, 4> <72, 1, *>
<5, 1, *, A> <8, 1, *> <48, 1, 4> <78, 1, *> 1, delete entry <48, 1, *>
<43, 1, *, B> <13, 1, *> <52, 1, 2> <83, 1, *> at timestamp 4.
<72, 1, *, C> <25, 1, 3> <59, 1, 3> <95, 1, 3>
<43, 1, *, C´> <27, 1, 3> <27, 1, 3> <99, 1, *> 2, B weak version
<39, 1, 3> <68, 1, 3> <102, 1, *> underflow since only
entry <43, 1, *> is alive.
3, copy live entries from
sibling node C to C´.
C´
<43, 4, *> 3, insert <43, 4, *> into
<72, 1, *> C´cause strong version
<78, 1, *> overflow.
<83, 1, *> 4, key split and node D
<99, 1, *> and E are created.
<102, 1, *>
13
 Mergence and key split
pversion =psvu=1/3
Root A B C
psvo=5/6
<5, 1, *> <43, 1, 4> <72, 1, *>
<5, 1, *, A>
<8, 1, *> <48, 1, 4> <78, 1, *> 1, delete entry <48, 1, *>
<43, 1, *, B>
<13, 1, *> <52, 1, 2> <83, 1, *> at timestamp 4.
<72, 1, *, C>
<25, 1, 3> <59, 1, 3> <95, 1, 3>
<43, 4, *, D> 2, B weak version
<27, 1, 3> <27, 1, 3> <99, 1, *>
<83, 4, *, E> underflow since only
<39, 1, 3> <68, 1, 3> <102, 1, *>
entry <43, 1, *> is alive.
3, copy live entries from
sibling node C to C´.
D E
3, insert <43, 4, *> into
C´cause strong version
<43, 4, *> <83, 1, *> overflow.
<72, 1, *> <99, 1, *>
<78, 1, *> <102, 1, *> 4, key split and node D
and E are created.

14
 Historical R-tree
• The structure maintains an R-tree for each
timestamp.
• Good for timestamp queries.
• Need a lot of space.

Example of an HR-tree 15
 3D R-tree
• Good for time interval queries

A timestamp query in 3D R-tree 16

 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

17
 MV3R-tree
• Intruduction
• MVR-tree
• Insertion and overflow handling in MVR-
tree
• Reinsertion
• Deletion and underflow handling in MVR-
tree
• Auxiliary 3D R-tree
• Query processing with MV3R-tree
18
 Introduction
• An idea to deal with both timestamp and interval queries.
• MVR-tree + 3DR-tree

Overview of an MV3R-tree
19
 MVR-Tree
• Multi-version R-tree.
• Can contain multiple R-trees.
• Each entry has the form <s, tstart, tend, pointer>

1, A to G are object boxes.

2, H, I, and J are leaf nodes.
3, C, I and K are alive
(unbounded).

A small MVR-tree with height 2

b=3
pversion=1/3 20
 Insertion and overflow handling in
MVR-tree
• Not overflow in intermediate nodes:
– Set tstart to the current time.
• Overflow in intermediate nodes:

Insertion in intermediate nodes

21
 Insertion and overflow handling in
MVR-tree
• Not overflow in leaf nodes:
– Set tstart to the current time.
• Overflow in leaf nodes:

Insertion in leaf nodess 22

 Reinsertion
• Any leaf node of MVR-trees can store a
reinserted entry.
• Different from MVB-tree.
• If general key split fails.

23
Deletion and underflow handling in
MVR-tree
• Not underflow in intermediate nodes:
– Modify tend from * to the current time.
• Underflow in intermediate nodes:
– Set tend of its live entries to current time.
• All entries are dead.
– Reinsert these entries to the most recent
logical R-tree after setting tstart=current time.
– MVB-tree handle underflows by merging with
sibling nodes.
24
 Deletion and underflow handling in
MVR-tree
• Not underflow in leaf nodes:
– Modify tend from * to the current time.
• Underflow in leaf nodes:
– First attempt to borrow a live entry from a sibling node.
– entry reinsertion if the heuristic fails (cause redundancy).
1, At timestamp 2, entry
A1 is deleted.
2, node A weak version
underflow.
3, version condition must
still be satisfied in B after
removal. (removal of B2
or B3 can cause weak
version underflow for
Borrowing a live entry from a sibling node b=8, 25
timestamp 1.)
Pversion=1/3
 Auxiliary 3D R-tree
• In order to process long time interval
queries.
• 3D R-tree is built on the leaves of the
MVR-tree.
• Whenever a leaf node of the MVR-tree is
updated, the change is propagated to its
entry in the 3D R-tree.

26
 Query processing with MV3R-tree

• Choose between 3D R-tree and MVR-tree.

– 3D R-tree is preferable for long time interval queries.
– MVR-tree is good for timestamp queries.
– Threshold is given for short time interval queries.
• When use MVR-tree, it choose the MVR-trees
whose roots´jurisdiction interval cover the
queried timestamp or interval.
• Duplicate visits for time interval queries.
– Duplicate data is created in version split or entry
reinsertion.
27
 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

28
 Strong and weak points
• Strong points:
– Appropriate related work and good structural
sequence.
– Good graphic explaination method.
• Weak points:
– No method to decide the parameters.
– Does not give a solution to deal with the
duplicate visit to the same node via different
parents in query processing with MVR-trees.
29
 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

30
 Relation to our project
• Similar work to our project but different
direction.

• MVB-tree is the extension of B-tree which

we use in our project.

• Many ideas can be used in our project.

31
 Overview
• Motivation
• Related works
• MV3R-tree
• Strong and weak points
• Relation to our project
• Conclusion

32
 Conclusion
• MV3R-tree can handle both timestamp
and interval queries efficiently.
• But update process is complex.
• Choose between MVR-tree and 3D R-tree
is important.
• Duplicate can only be reduced but not be
avoided.

33