GMOBench: Benchmarking generic moving objects

Abstract

In real world scenarios, people’s movement include several environments rather than one, for example, road network, pavement areas and indoor. This imposes a new challenge for moving objects database that the complete trip needs to be managed by a database system. In the meantime, novel queries regarding different transportation modes should also be supported. Since existing methods are limited to trips in a single environment and do not support queries on moving objects with different transportation modes, new technologies are essentially needed in a database system. In this paper, we introduce a benchmark called GMOBench that aims to evaluate the performance of a database system managing moving objects in different environments. GMOBench is settled in a realistic scenario and is comprised of three components: (1) a data generator with the capability of creating a scalable set of trips representing the complete movement of humans (both indoor and outdoor); (2) a set of carefully designed and benchmark queries; (3) Mode-RTree, an index structure for managing generic moving objects. The generator defines some parameters so that users can control the characteristics of results. We create the benchmark data in such a way that the dataset can mirror important characteristics and real world distributions of human mobility. Efficient access methods and optimization techniques are developed for query processing. In particular, we propose an index structure called Mode-RTree to manage moving objects in different environments. By employing the proposed index, the cost of benchmark queries is greatly reduced. GMOBench is implemented in a real database system to have a practical result. We perform an extensive experimental study on comprehensive datasets to evaluate the performance. The results show that by using the Mode-RTree we achieve significant performance improvement over the baseline method, demonstrating the effectiveness and efficiency of our approaches.

Appendices

Appendix A - Formulate Benchmark queries

To formulate queries, we provide a relational interface to exchange information. Several relations are provided to manage infrastructure data and moving objects, summarized in Table 4. $\underline {route}$ is a data type that we propose to represent bus (metro) routes. A bus (metro) route is represented by a sequence of segments each of which defines the locations of start and end bus stops as well as the connection described by a line. For the indoor environment, we design a data type called $\underline {groom}$ used in Rel_Room to represent all rooms of a building, e.g., office rooms, corridors, staircases. Since in some cases one room can have several floors such as an amphitheater and a chamber, a $\underline {groom}$ object consists of a set of elements each of which defines the 2D area of a floor and the height above the ground level.

Table 4 Data relation schemas

To access an infrastructure, we assign a symbol to each relation (summarized in Fig.12a) and obtain the data from the operator get_infra described in Fig. 12b. Figure 13 lists the operators that are used to access the data and formulate the queries. To represent the trajectories of moving objects (i.e., the projection into space), we propose a data type $\underline {genrange}$ which defines a set of elements. Each element records the path according to a referenced object as well as the transportation mode.

Fig. 12

Access infrastructures

Fig. 13

Operators used in queries

We use qt to denote the query time parameter. For the queries (Q4, Q5, Q17, Q18, Q20) on both bus and metro, it is sufficient to show the query expression for a bus network as the procedure is similar for metros.

• Q1. Find out all metros passing through the city center.

  Let Reg_C be a region denoting the city center area.

  • SELECT mr.Mr_id

  • FROM get_infra (SpaceGendon, METROROUTE) as mr

  • WHERE mr.GeoData intersects Reg_C

• Q2. Given a building named by X, find all bus stops within 300 meters.

  • SELECT bs FROM get_infra (SpaceGendon, BUILDING) as b,

  • get_infra (SpaceGendon, BUSSTOP) as bs WHERE b.Name = X and distance (b.GeoData, bs.GeoData) < 300

• Q3. Which streets does bus No. 12 pass by?

  • SELECT r.Name

  • FROM get_infra (SpaceGendon, BUSROUTE) as br

  •     get_infra (SpaceGendon, ROAD) as r WHERE br.Br_id = 12 and br.GeoData intersects r.GeoData

• Q4. Where can I switch between bus route No. 16 and No. 38?

  • SELECT bs1, bs2

  • FROM get_infra (SpaceGendon, BUSSTOP) as bs1,

  •     get_infra (SpaceGendon, BUSSTOP) as bs2, WHERE bs1.Br_id = 16 AND bs2.Br_id = 38 AND bs1.GeoData = bs2.GeoData

• Q5. At 8am on Monday, who sits in the bus No. 32?

  • SELECT mo.Name

  • FROM get_infra (SpaceGendon, BUS) as bus, MOGendon as mo

  • WHERE ref_id(val (mo.Traj atinstant qt)) = bus.Bus_id and bus.Br_id = 32

• Q6. What is the percentage of people traveling by public transportation vehicles?

  • Let no_mo = SELECT COUNT(⋆) FROM MOGendon

  • Let no_submo = SELECT COUNT(⋆)

  •         FROM MOGendon as mo

  •         WHERE mo.Traj contains BUS or

  •            mo.Traj contains METRO or

  •            mo.Traj contains TAXI

  • SELECT DIV(no_submo, no_mo)

• Q7. Where and how long does Bobby walk during his trip?

  • SELECT mo.Traj at Walk

  • FROM MOGendon as mo

  • WHERE mo.Name = "Bobby"

  • SELECT duration(deftime (mo.Traj at Walk))

  • FROM MOGendon as mo

  • WHERE mo.Name = "Bobby"

• Q8. Find out all people passing room 312 at the office building between 9:00am and 11:00am on Monday.

  • SELECT mo.Name

  • FROM MOGendon as mo,

  •      get_infra (SpaceGendon, BUILDING) as b,

  •      get_infra (SpaceGendon, ROOM) as rm

  • WHERE b.Name = "Office-X" and b.B_id = rm.B_id and

  •      and rm.Room_id = 312 and

  •      ((mo.Traj at Indoor) atperiods qt) contains rm.Room_id

• Q9. Who arrived by taxi at the university on Friday?

  To arrive by taxi at the university means that the final location of the passenger within the taxi belongs to some driveway area close to the university. Such a part of the road network is also managed in the database as an object UniDriveway of type line.

  • SELECT mo.Name

  • FROM MOGendon AS mo

  • WHERE val(final ((mo.Traj atperiods qt) at Taxi)))

  •      inside UniDriveway

• Q10. Who entered bus No. 3 at bus stop University on Tuesday afternoon?

  • SELECT mo.Name

  • FROM MOGendon AS mo,

  •      get_infra (SpaceGendon, BUSSTOP) AS bs,

  •      get_infra (SpaceGendon, BUS) AS bus

  • WHERE bs.Br_id = 3 AND bs.Name = "University" AND bus.Br_id = 3 AND

  •       bs.GeoData = freespace(val(initial ((mo.Traj atperiods qt) at

  •                theloc (bus.Bus_id, undef, undef)))))

• Q11. Find out all people walking through zone A and zone B on Saturday between 10am and 3pm.

  • SELECT mo.Name

  • FROM MOGendon AS mo,

  •      get_infra (SpaceGendon, RBO) AS R1,

  •      get_infra (SpaceGendon, RBO) AS R2

  • WHERE R1.Name = "Zone-A" AND R2.Name = "Zone-B" AND

  •        ((mo.Traj atperiods qt) at Walk) passes R1.Reg AND

  •        ((mo.Traj atperiods qt) at Walk) passes R2.Reg

• Q12. Did anyone who was on floor H-5 of the office building between 2pm and 5pm take a bus to the stop “train station” on Friday?

  To be on floor H-5 means to be in any of the rooms of floor H-5. We create a table associating rooms with their floors:

  Uni_Rel(B_id: int, Floor: string, Room_id: int)

  • SELECT mo.Name

  • FROM MOGendon AS mo,

  •       Uni_Rel AS u,

  •       get_infra(SpaceGendon, ROOM) AS r,

  •       get_infra(SpaceGendon, BUSSTOP) AS bs1,

  •       get_infra(SpaceGendon, BUSSTOP) AS bs2

  • WHERE u.Floor = "H-2" AND u.B_id = r.B_id AND u.Room_id = r.Room_id AND

  •       (mo.Traj atperiods qt) passes r.GeoData AND

  •       bs1.Name = "University" AND

  •       bs2.Name = "Main station" AND

  •       freespace(val(initial ((mo.Traj atperiods qt) at Bus))) =

  •       bs1.GeoData AND

  • freespace(val(final ((mo.Traj atperiods qt) at Bus))) =

  • bs2.GeoData

• Q13. Did bus No. 35 pass any bicycle traveler by on Monday?

  We define pass to mean that there exists a time instant that the distance between the two moving objects is less than 3 meters.

  • SELECT mo.Name

  • FROM get_infra (SpaceGendon, BUS) AS bus,

  •       MOGendon AS mo

  • WHERE bus.Br_id = 35 AND

  •       sometimes(distance(freespace (bus.Traj atperiods qt),

  •                      freespace ((mo.Traj atperiods qt) at Bicycle)) < 3.0)

• Q14. Did someone spend more than 15 minutes on waiting for the bus at the bus stop Cinema on Saturday?

  • SELECT mo.Name

  • FROM MOGendon AS mo,

  •       get_infra (SpaceGendon, BUSSTOP) AS bs

  • WHERE bs.Name = "Cinema" AND

  •       duration(deftime ((mo.Traj atperiods qt) at bs.GeoData)) > 15

• Q15. How many people visit the cinema on Saturday?

  • SELECT COUNT(⋆)

  • FROM MOGendon AS mo,

  •       get_infra (SpaceGendon, BUILDING) AS b

  • WHERE b.Name = "Cinema" AND

  •       (mo.Traj atperiods qt) contains b.B_id

• Q16. Find out all people staying at room 154 in the university for more than one hour on Thursday.

  • SELECT mo.Name

  • FROM MOGendon AS mo,

  •       get_infra (SpaceGendon, BUILDING) AS b,

  •       get_infra (SpaceGendon, ROOM) AS r

  • WHERE b.Name = "University" AND b.B_id = r.B_id AND

  •       r.Room_id = 154 AND

  •       duration(deftime (mo.Traj atperiods qt) at

  •            theloc (r.Room_id, undef, undef))) > 60

• Q17. Did someone spend more than one hour traveling by bus?

  • SELECT mo.Name

  • FROM MOGendon AS mo

  • WHERE duration(deftime (mo.Traj at Bus)) > 60

• Q18. Find out all people changing from bus No. A to bus No. B at stop X.

  • SELECT mo.Name

  • FROM MOGendon AS mo

  •       get_infra (SpaceGendon, BUSSTOP) as bs,

  •       get_infra (SpaceGendon, BUS) as bus1,

  •       get_infra (SpaceGendon, BUS) as bus2

  • WHERE bus1.Br_id = A AND bus2.Br_id = B AND

  •       bs.Name = X AND

  •       freespace(val(final (mo.Traj at theloc(bus1.Bus_id)))) =

  • freespace(val(initial (mo.Traj at theloc(bus2.Bus_id)))) AND

  • freespace(val(final (mo.Traj at theloc(bus1.Bus_id)))) =

  •       bs.GeoData

• Q19. Find out all trips starting from zone A and ending in zone B by public transportation vehicles with the length of the walking path being less than 300 meters.

  • SELECT mo.Name

  • FROM MOGendon AS mo

  •       get_infra (SpaceGendon, RBO) AS R1,

  •       get_infra (SpaceGendon, RBO) AS R2

  • WHERE R1.Name = "ZoneA" AND R2.Name = "ZoneB" AND

  •       val(initial (mo.Traj)) inside R1.GeoData AND

  •       val(final (mo.Traj)) inside R2.GeoData AND

  •       ((mo.Traj contains Bus) OR (mo.Traj contains Metro) OR

  •       (mo.Traj contains Taxi)) AND

  •       length(trajectory (mo.Traj at Walk)) < 300

• Q20. Find the top k bus (metro) routes with high passenger flow for all workdays.

  • SELECT TOP k ⋆

  • FROM

  •       SELECT bus.Br_id, COUNT(⋆) AS NO

  •       FROM MOGendon AS mo

  •            get_infra (SpaceGendon, BUS) AS bus

  •       WHERE mo.Traj contains Bus AND

  •            ((mo.Traj atperiods qt) at Bus)) contains bus.Bus_id

  •       GROUP BY bus.Br_ide

  •       ORDER BY NO DESC

• Q21. Find the top k road segments with high traffic during the rush hour for all workdays.

  To answer such a query, we create a relation storing all road segments with the schema

  Rel_RoadSeg: (S_id: int, GeoData: line, R_id: int)where S_id is the unique segment id, GeoData stores the geometrical property and R_id indicates the id for the road that the segment belongs to. For each road, we decompose it into a set of pieces at the junction positions and each piece is stored as a tuple in Rel_RoadSeg.

  First, we get the traffic of each road segment by aggregrating the number of buses passing by.

  • LET Rel_BRCOUNT =

  •       SELECT br.Br_id, br.GeoData, COUNT(⋆) AS NO

  •       FROM get_infra (SpaceGendon, BUSROUTE) as br,

  •            get_infra (SpaceGendon, BUS) as bus

  •       WHERE deftime (bus.Traj) intersects qt AND bus.Br_id = br.Br_id

  •       GROUP BY br.Br_id

  Each tuple in Rel_BRCOUNT indicates the number of trips for each bus route.

  • LET Rel_RES1 =

  •       SELECT s.S_id, SUM(br.NO) AS FLOW

  •       FROM Rel_BRCOUNT as br,

  •            Rel_RoadSeg as s

  •       WHERE br.GeoData intersects s.GeoData

  •       GROUP BY s.S_id

  We perform the join on bus routes and road segments to get the segments that each bus route maps to. Then, we group the tuple by segment id and call the aggregation function to get the total number of buses passing a segment from different routes.

  Second, we get the traffic from moving objects that contain one of the modes {Car, Taxi, Bike }. At first, we collect the paths of these trips.

  • LET Rel_Traj_C =

  •       SELECT trajectory (mo.Traj at CAR) AS Path

  •       FROM MOGendon AS mo

  •       WHERE deftime (mo.Traj) intersects qt

  • LET Rel_Traj_T =

  •       SELECT trajectory (mo.Traj at TAXI) AS Path

  •       FROM MOGendon AS mo

  •       WHERE deftime (mo.Traj) intersects qt

  • LET Rel_Traj_B =

  •       SELECT trajectory (mo.Traj at BIKE) AS Path

  •       FROM MOGendon AS mo

  •       WHERE deftime (mo.Traj) intersects qt

  • LET Rel_Traj = Rel_Traj_C union Rel_Traj_T union Rel_Traj_B

  Then, we do the join on the paths and road segments to get the traffic value.

  • LET Rel_RES2 =

  •       SELECT s.S_id, COUNT(⋆) AS FLOW

  •       FROM Rel_Traj AS p,

  •            Rel_RoadSeg AS s

  •       WHERE p.Path intersects s.GeoData

  •       GROUP BY s.S_id

  Third, we merge the two traffic relations Res_RES1 and Res_RES2 to get the final result.

  • SELECT TOP k ⋆

  • FROM

  •      SELECT r1.S_id, r1.FLOW + r2.FLOW as C

  •      FROM Rel_RES1 as r1,

  •          Rel_RES2 as r2

  •      WHERE r1.S_id = r2.S_id

  •      ORDER BY C DESC

Appendix B

1.1 Unique IFOB id and reference id for an indoor location

In order to have a unique id for each IFOB, we define a set of disjoint ranges each of which is used for one infrastructure, e.g., [1, 5000] for I _rn , [6000, 7000] for I _bn . Given a generic location $gl \in D_{\underline {genloc}}$, the meaning of gl.oid is clear if the IFOB belongs to an outdoor infrastructure. However, for the indoor environment different buildings can have the same room id (e.g., ROOM NO 12, ROOM NO 35), only using the building id cannot uniquely identify an indoor location.

To solve the problem, gl.oid is set by combining a building id and a room id for an indoor location. We introduce how to achieve the goal in the following. For the sake of clear presentation, we use two strings B_Id and R_Id to denote a building id and a room id, respectively. Suppose that gl is located in the room R_Id = “123” of a building B_Id = “167654”, then gl.oid is assigned by the concatenation of B_Id and R_Id, that is “167654123”. Let len(B_Id) return the length of a string for B_Id. The range for all building ids is carefully chosen in order to fulfill the condition len(B_Id _min )=len(B_Id _max ), i.e., the number of digits is the same for each id. Otherwise, an ambiguous case can occur.

Afterwards, len(B_Id) is a fixed value (new building construction is not considered) and we can extract B_Id and R_Id from an indoor location. Using the example before, we can get B_Id = “167654” and R_Id = “123“ from gl.oid = “167654123”.

Appendix C - experimental statistics

Fig. 14

Scaling datasets by employing the Mode-RTree

Fig. 15

Scaling datasets by baseline method

Fig. 16

CPU time (sec) for B2M

Fig. 17

I/O accesses (k) for B2M

Fig. 18

CPU time (sec) for H1M

Fig. 19

I/O accesses (k) for H1M

Fig. 20

Query parameters: B2M

Fig. 21

Query parameters: H1M