On Scheduling Real-time Multi-item Queries in Multi-RSU Vehicular Ad Hoc Networks (VANETs) G. G. Md. Nawaz Ali City University of Hong Kong Kowloon, Hong Kong Email: [email protected]

Edward Chan City University of Hong Kong Kowloon, Hong Kong Email: [email protected]

Abstract—The issue of multi-item queries in wireless broadcasting systems has received considerable interest recently. Two problems, namely query starvation and bandwidth utilization, have been identified as key issues that need to be solved. In this paper, we examine this problem in the context of VANETs with multiple cooperating Road Side Units (RSUs). We characterize a query with two deadlines: query total deadline (QTD) which is the actual deadline of a query and query local deadline (QLD) which is the duration a vehicle dwells in an RSU range after submitting that query. By careful consideration of these deadlines, vehicle speed, RSU range and inter-RSU distance, we propose a Cooperative Query Serving (CQS) approach which allows multiple RSUs to share residual bandwidth, deal effectively with both the query starvation and the bandwidth utilization problem and hence maximize the chance of serving multiple items queries. Simulation results show CQS outperforms other scheduling algorithms. Keywords-VANETs; Roadside-to-vehicle Communication (RVC); Road Side Units (RSUs); real-time scheduling; multi-item queries.

I. I NTRODUCTION A number of applications have been envisioned in the Intelligent Transportation System (ITS) such as road safety, driving assistance, internet service from on board vehicles etc. Generally vehicle-to-vehicle (V2V) and vehicle-toinfrastructure (also called Road Side Units (RSU)) (V2I) are the two communication models in ITS. Our focus in this paper is the second of these two communication models. The US Federal Communications Commission (FCC) has formulated the Dedicated Short Range Communication (DSRC) specifications for VANETs. In DSRC, 75 MHz is split into seven channels, namely six service channels (SCH) or data channels (DCH) and one control channel (CCH) each with 10 MHz band. However multi-channel operation in IEEE 802.11 radio is extremely challenging, since current hardware can demodulate one channel at a time [1]. A dual channel RVC (Roadside-to-vehicle Communication) model is proposed by Maeshima et al. [2], one CCH operates in the 5.8 GHz band of DSRC for safety message and one DCH uses VHF/UHF band for safety and non-safety message, in where with a flexibility that the DCH can be converted for safety message communication in needs. Mak et al. [1] propose an RVC system, where an RSU and a vehicle are equipped with two dedicated 802.11 DSRC radio channels: one CCH and one DCH. An RSU, through the CCH, can announce the upcoming data transfer, delivers

Wenzhong Li Nanjing University Nanjing, China Email: [email protected]

safety message and received request from vehicles, whilst data transfer takes place in the DCH. Traditionally, queries in VANETs are assumed to request only a single data item per query. However, in many applications, a generated query can request for multiple data items, which is termed called multi-item query. Examples include a query for real-time traffic condition of several alternative routes towards a destination, or making a trading decision after knowing the market price of several companies for a same product. The primary difference from single item query is that a multi-item query can only be satisfied if all the queried data items are served. Achieving good performance for multi-item queries is the subject of a number of studies in recent years. Chung et al. [3] study multi-item query in push based broadcast environment with fixed duration broadcasting cycle; they propose different techniques to place the queried data items in that cycle in order to minimize the access time. N. Prabhu and V. Kumar [4] use indexing techniques to minimize the access time and tuning time for on-demand broadcasting. However all these works are based on non-real-time and fixed sized data items. Chen et al. [5] and Liu et al. [6] also study the multi-item queries for real-time on-demand broadcasting environment. However they consider only fixed sized data item and a single server environment. For making the query real-time they assign a deadline of a submitted query where beyond that deadline any unserved or partially served query termed as deadline missed. To minimize the deadline miss ratio (the number of deadline miss queries over total queries) they propose an approach which balances the trade-off between two issues: query starvation problem and bandwidth utilization problem. Typically a broadcast system will maximize the use of the available bandwidth by broadcasting the most popular data items; however this mean that a query may need to wait for an excessively long time if one of its requested data items is a cold data item (also called unpopular data item) i.e. the query will starve. The challenge is to reconcile these two conflicting requirements. In this paper, we consider the multiple-item query problem in the context of a VANETs infrastructure based on multiple cooperating Road Side Unit (RSUs), with particular emphasis on real-time queries. The objectives of this paper are, firstly, introducing non-safety multi-item queries in RVC system where queried data items can be diverse in size, and secondly, through efficient scheduling approach, properly

Interconnected RSUs

Database

Scheduler Received queue

RSU service range

Q1, Q3 Q4, Q2 Q10, Q7, Q9 Q6, Q8, Q5 ..

Control Channel (CCH)

Scheduled data item for serving d1 t1 d4 t2 d2 t3 dn

tn

Data Channel (DCH)

RSU

Figure 1. model

System architecture of cooperative multi-item queries serving

utilize the data channels (DCHs) bandwidth of networked multiple RSUs and to maximize the overall system performance. II. S YSTEM M ODEL A. System Architecture Similar to [1], in our proposed RVC system, each RSU consists of separate DSRC CCH and one DCH. Through the CCH an RSU delivers safety messages periodically and index information of the next broadcast for non-safety message, so that an passing vehicle tunes its DCH with the DCH of RSU, if needs that information. Safety message is smaller in terms of size and amount, and hence consume less bandwidth [1]. Hence, in this paper we only focus on the scheduling of the non-safety message in the DCH. The system architecture of our proposed model is shown in Fig. 1. Upon entering into the RSU range, a vehicle can generate queries for non-safety messages through its CCH, and receive responses through the DCH until it leaves the range of the RSU. As RSUs are interconnected, they can transfer partially or fully unsatisfied queries to their neighbor RSUs. An RSU invokes the underlying scheduling algorithm to select a query for serving in the next cycle. If a query is transferred, it will be considered for scheduling after the corresponding vehicle reaches the transmission range of transferee RSU. In our proposed cooperative multiple RSUs model, if a submitted query is not completely satisfied within the dwelling time of the corresponding vehicle at an RSU, the unserved query will be then transferred to the other RSU which is located in the direction of the corresponding vehicle is heading. Upon reaching at the transferee RSU, the vehicle will issue an inform query through the control channel, where the inform query is much smaller in size (may only contain vehicle ID and query ID) in comparison of normal submitted query, which helps to save the scare bandwidth of the control channel. B. Notations and Assumptions Definition 1: RSU. Assume an RSU database has a total of DBSIZE data items which are read-only and readily

available upon request. A data item is denoted by di where 1 ≤ i ≤ DBSIZE and size of di is S(di ). P(di (t)) denotes the popularity of the data item di at time t, that is the number of outstanding queries for data item di at time t at an RSU received queue. The radius of an RSU transmission range is R and channel bandwidth is CBW . Hence the service time of a data item di S(di ) is: Tdserv = CBW . The average velocity of a vehicle inside an i RSU transmission range is V, hence the maximum dwelling dwll = 2R . The interperiod of a vehicle at an RSU is: Tmax V RSU distance a vehicle travels outside the RSU transmission range is called the blind zone, and per hop blind zone travel travel = davg , where average blind zone distance is time is, Thop V davg . Definition 2: Query. A vehicle submitted query Qi at RSUi is characterized by the following tuples: D Qi = (V hi , d(Qi ), Tqgnr , TqQT , HD(V hi )) i i

Here V hi : Vehicle id which generates query Qi ; d(Qi ): Requested data items set of query Qi , which can be explored as an un-ordered data items set, d(Qi ) = d1 (qi ), d2 (qi ), · · · dQL(Qi ) (qi ), where for a queried data item d j (qi ), 1 ≤ j ≤ DBSIZE. QL(Qi ) denotes the query length of the query Qi , namely the number of data items requested by a multi-item query Qi . d ′ (Qi ) denotes the set of unsatisfied data items of query Qi , where d ′ (Qi ) = ′ ′ d1′ (qi ), d2′ (qi ), dQL ′ (Q ) (qi ) and QL (Qi ) ≤ QL(Qi ).Qt (RSUi ) i denotes the set of submitted queries pending at RSUi at time t. serv = Total serving time of query Qi of length QL is, TQL(Q i) serv ∑∀di ∈d(Qi ) Tdi . Similarly unserved serving time of query Qi serv serv ; is, TQL ′ (Q ) = ∑∀d ′ ∈d ′ (Qi ) Td ′ i i i gnr Tqi : At time when query Qi is generated, its value can be dwll (V h )); Tqgnr = random(0.0, Tmax i i QT D Tqi : Assigned total deadline value of the query Qi . After expiring this value Qi will be termed as invalid query and discarded from the RSU received queue and then regarded D as deadline missed query. Assigned TqQT value by vehicle i V hi must be equal or greater than the serving time of the requested data item set d(Qi ). Again, V hi must stay the dwelling period Tvhdwll (RSUi ) at RSUi after generating Qi . i QT D Hence Tqi will be: serv D TqQT ≥ max{TQL(Q , Tvhdwll (RSUi )} i i i)

(1)

HD(V hi ): Direction in which vehicle is heading V hi . HD(V hi ) could be North (N), South (S), East (E) and West (W ). Definition 3: Query Deadline. A submitted query Qi carries two deadlines: 1) Query Total Deadline (QTD): QTD is a query Qi D assigned total deadline value TqQT , within this time i period Qi must be satisfied completely otherwise it will be considered having missed its deadline. 2) Query Local Deadline (QLD): The time period (TqQLD ) i a vehicle dwells within the range of RSUi after generating query Qi or dwells in RSU j range if Qi is

RSUi

RSUk

RSUj

R dRSUi,j Blind zone distance

N W

E Remaining QTD=

S TqiQLD(RSUj) = 2R/V

TqiQLD(RSUi) =2R/V-Tqignr

TqiQTD- {TqiQLD(RSUi)+T RSUi,j bz_travel+TqiQLD(RSUj)}

T RSUi,j bz_travel

If after dwelling TqQLD (RSU j ) time at RSU j , all of the i unserved data items in d ′ (Qi ) are served, query Qi is treated as satisfied. On the other hand, if all the data items in d ′ (Qi ) cannot satisfy within time period TqQLD (RSU j ), the i next transfer of Qi will depend on its remaining QTD value and number of unserved data items in d ′ (Qi ). At time t a query Qi becomes infeasible for serving if its remaining QTD value is smaller than required serving time of its unserved data item, that is: D serv (TqQT − t) < TQL ′ (Q ) i i

TqiQTD = QTD

(5)

Time

Figure 2.

Calculating QLD value from QTD

transferred in there. If Qi is transferred to another RSU j , its QLD value will be recalculated from its remaining QTD value. Calculating QLD: From a query Qi , an RSUi can estimate the dwelling period of a vehicle V hi . From Fig. 2, the QLD of a query Qi generated at time Tqgnr at RSUi is: i TqQLD (RSUi ) = Tvhdwll (RSUi ) = i i

2R − Tqgnr i V

(2)

TqQLD (RSUi ) could be smaller than the required period for i serving all the data items in the set d(Qi ). In that case, Qi cannot be satisfied completely at RSUi , hence to serve the unsatisfied data items in the set d(Qi ), Qi needs to transfer to an another RSU j which can satisfy the query. As time passes, the query local deadline, TqQLD (RSUi ) i decreases and becomes zero when V hi leaves the RSU range. If Qi is not satisfied completely within TqQLD (RSUi ), before i transferring to another RSU j (say RSU j with its own load has the spare serving capacity of query Qi ), RSUi does the feasibility check of query Qi . Query Qi is only feasible for transferring to RSU j if it has enough remaining QTD value D′ (TqQT ) at time t to serve all of the unserved data item in i d ′ (Qi ), that is, D′ TqQT i

D = (TqQT − t) ≥ {TqQLD (RSUi ) + (nhops − 1)× i i serv dwll bz travel Tmax + TRSU + TQL ′ (Q ) } i, j i

(3)

bz travel where TRSU is the total blind zone travelling time of i, j vehicle V hi from RSUi to RSU j which is calculated as, bz travel travel , here assumed RSU and RSU are TRSU = nhops × Thop i j i, j nhops apart. If a partially or completely unsatisfied feasible query Qi of vehicle V hi is transferred to an another RSU j , it gets a new QLD value there. Transferred query Qi becomes active after V hi arrives at RSU j . Upon arriving V hi at time t, RSU j assigns a new QLD value to Qi . This time QLD of Qi depends on its remaining QTD value. If its remaining QTD value is smaller than maximum dwell time value ( 2R V ), it gets the remaining QTD value else maximum dwell time value as its QLD value. Hence at time t, Qi will get its new QLD value at transferee RSU j is:

TqQLD (RSU j ) = min{ i

2R D , (TqQT − t)} i V

(4)

An infeasible query is one which cannot be satisfied completely within in its remaining QTD value and will be removed from the respective sub-queue. Our target is to propose an efficient approach which can satisfy as many data items as possible in d(Qi ) of a query (RSUi )) and Qi within its dwelling period at RSUi (TqQLD i satisfy the rest of the unserved data items (if there is) in d ′ (Qi ) cooperatively with the help of neighbor RSUs within Q′i s remaining QTD value. III. P ROPOSED S OLUTION We summarize our proposed Cooperative Query Serving (CQS) scheme for handling multi-item queries as follows: Step 1. Calculate QLD value of the queries: QLD value calculation of a query is done using equation 2 if it is RSU’s own query and using equation 4 if it is received from other RSUs (transferred query). Step 2. Assign received queries in the respective subqueue: In the cooperative load balancing approach a partially or completely unsatisfied query Qi has a chance to satisfy from a neighbor RSU at least one hop away if it has enough remaining QTD value at time t, that is: D travel serv (TqQT − t) ≥ {(TqQLD (RSUi ) + Thop + TQL ′ (Q ) } i i i

(6)

However the query which does not have enough remaining QTD value to satisfy the equation 6, RSUi is the last resort for them to be satisfied completely. Such type queries can be RSUi ’s own queries or transferred from its neighbor RSU from other RSUs. Hence this type queries should get higher priority over the queries which have enough remaining QTD value to satisfy equation 6. For this we assign the RSUi received queries into two sub-queues at time t, which are: QtNCQ (RSUi ) (Non-critical sub-queue): The query Qi which remaining QTD value at time t satisfies the condition 6. QCQ t (RSUi ) (Critical sub-queue): The query Qi which doesn’t have enough remaining QTD value at time t to satisfy the condition 6. Step 3. Query selection: We select a query Qi for service from the critical subqueue QCQ t (RSUi ) based on the remaining QLD value at time t (TqQLD (RSUi ) −t), the unserved query length QL′ (Qi ) i serv . Query selection is and the unserved serving time TQL ′ (Q ) i performed based on the following criteria:

(1) For two queries having the same QL′ value (same unserved query length), select the one having the lower remaining QLD value (the most urgent one). (2) For two queries with the same remaining QLD value, select one having lower QL′ value. This ensures the query having lowest number of data items unserved will get the chance to be served completely. (3) For two queries having the same remaining QLD and serv QL′ value, select one with the lower TQL ′ (Q ) value, since i this helps the query needing a smaller service time to be completed. The last two criteria ensure a query does not have to wait a long time for serving a queried cold data item and increase the chance that a query that is close to completion can be satisfied, which is the key to alleviate the query starvation problem. Hence query Qi is selected from the sub-queue QCQ t (RSUi ) which has the lowest Query Selection Value(Qi (t)) value at time t, where: Query Selection Value(Qi (t)) = {(TqQLD (RSUi ) − t)+ i serv ′ TQL ′ (Q ) } ∗ QL (Qi ) i

Step 4. Data item selection from the selected query: For service in the next service cycle, a data item di of the selected query Qi is selected from the unserved set of d ′ (Qi ) which has the maximum popularity at time t, that is, di = max{P(d ′j (t))|∀d ′j ∈ d ′ (Qi )}. As the selected data item di is the most popular one among the unserved requested data items of query Qi , there is a high chance di is also requested by other queries residing in either sub-queues. Hence serving di addresses the bandwidth maximization problem. Step 5. Update query status and query length: The queries residing in either the sub-queues QCQ t (RSUi ) or QtNCQ (RSUi ) request for the data item di , update their unserved data item set d ′ (Qi ) and query length QL′ (Qi ). If the QL′ (Qi ) of query Qi becomes zero, Qi is treated as completely satisfied and removed from the respective subqueue. Step 6. Query feasibility check: Query feasibility checking is done following equation 5, and removed the infeasible query accordingly. If QCQ t (RSUi ) sub-queue is empty, query selection (from Step 3) and onwards process goes on from sub-queue QtNCQ (RSUi ). Step 7. Query transfer from sub-queue QtNCQ (RSUi ): If the QLD value of a query Qi in QtNCQ (RSUi ) has expired but the query still have some unserved data items in d ′ (Qi ), the query transfer process is started. Query transfer of a query Qi is done based on the direction of vehicle V hi , the remaining QTD value of Qi and the available workload handling capacity of the target transferee RSUs in that direction. The transferring procedure of a query Qi from RSUi to RSU j (one of the RSUs in the neighbor RSU set to a specific direction of RSUi ) will not be described in detail here due

Table I S IMULATION PARAMETERS Parameter Default Range NRSU 27 — Nvehicle 45 15-75 Poisson mean (µ ) — 5-20 QSize 2-5 1 to 4-8 VGIV 1.50 — RGIV 1.00 — DBSIZE 600 — SIZEMIN 10 — SIZEMAX 500 — CBW 625 — RSU Range (R) 120 — V 40 10-70 Distance 100 — THETA (θ ) 0.8 — X, Y 4,3 2-6, 1-5 ε 10, 5.5 — TimeWindow (τ ) 5 —

Description Number of RSUs in the simulation topology Generated number of vehicles Mean number of requests generated by a vehicle at each RSU Query Size (Number of data items requested by a query) Vehicle generation interval (Exponential distribution) Request generation interval (Exponential distribution) Number of data items in the database Minimum size of requested data item (K bytes) Maximum size of requested data item (K bytes) Channel broadcast bandwidth (K bytes/s) (5 Mbps = 625 Kbytes) RSU communication range (m) Vehicle average speed (km/h) Inter RSU distance (m) Zipf distribution parameter Minimum and maximum laxity for calculating QTD value Constant used for calculating QLD value upon a query generation Duration of time window for calculating EWMA (sec)

to space limitation; a full description is provided in [7]. The key steps can be summarized briefly as follows: 1. If an RSU j is found from the neighbor RSUs set of RSUi to a specific heading direction which is capable to serve query Qi , transfer the query to RSU j . 2. Otherwise, find an RSUk from the neighbor RSU set of RSUi in that direction which will have minimum estimated workload (minimum Nusr (RSUk )) when V hi arrives there, and then transfer Qi to RSUk . IV. P ERFORMANCE E VALUATION A. Details of the Simulation For performance comparison against our proposed CQS approach we adopt the First Come First Served (FCFS) as well as the Most Request First (MRF) scheduling algorithms for handling multi-item queries in multiple RSUs based VANETs, and the following metrics for performance evaluation: (1) Deadline Miss Ratio (DMR): The number of multiitem queries missing the deadline to the total number of queries received by the system. Note that even if all but one item remains unsatisfied of a query before its deadline (QTD value) expires, the query is considered to have missed its deadline. (2) Success Rate of Transferred Queries (SRTQ): The ratio of the number of transferred queries (partially or completely unserved queries) successfully served before the query deadline (QTD value) to the total number of transferred queries. (3) Scheduling Efficiency Ratio (SER): SER is defined as the ratio of number of data items served which were queried by the satisfied queries to the total number of data items served [5], that is: SER =

# o f served data items in the satis f ied queries Total number o f served data items

Note that SER for multi-item queries is different from that of single-item queries because a multi-item query is successful only when all of its queried data items are served.

B. Simulation Model The default parameters we use for the experiments are shown in Table I. Our simulation environment is set up based on a real road scenario in an area sized 1.8 km × 1.8 km in Hong Kong. We generate vehicles from different generation points and let the vehicles move. When a vehicle reaches the end of its route, it is removed from the simulation. Vehicle Generation InterVal (VGIV) from each generation point follows exponential distribution. Vehicle mobility in the simulation region follows the Manhattan mobility model [8]. A vehicle can generate queries until it leaves the RSU transmission range. The QLD value of a generated query is calculated as follows: 2R serv dwll ) − Tqgnr )) , (Tmax (= TqQLD (RSUi ) = random(ε ∗ TQL(Q i i i) V serv TQL(Q varies from one query to another query because of i) different query length and queried data item size. The default minimum and maximum laxity for generating query length is 2 and 5 respectively. For assigning different QTD values to different generated queries we use the following formula: dwll dwll travel D = random(Tmax , (X ∗ Tmax +Y ∗ Thop )) TqQT i

(7)

dwll and (X ∗ T dwll + Y ∗ T travel ) are the minimum Here, Tmax max hop and maximum range respectively for calculating QTD value of a query. For varying upper limit of QTD value, we vary the maximum range value by varying the value of X and Y. The default values of X and Y are 4 and 3 respectively. Query generation follows the Poisson process, with the Poisson mean request generation µ value ranging from 5 to 20 for each vehicle. The vehicle request generation interval is exponentially distributed with a default mean value of 1.00. The request access pattern for accessing the data item in the RSU DATABASE follows the Zipf distribution. For generating diverse sized data items in the RSU database we use the random (RAND) data item size distribution The simulation completes when all the generated vehicles exit the simulated topology. For performance analysis we take the average data of 20 simulation runs for same simulation parameter settings with different seed values.

C. Performance Analysis We will compare the performance of CQS with other approaches in this section, both for stand-alone and for the scenario where cooperative load transfer (CLT) is available i.e. enabling load transfer among multiple RSUs. Impact of number of vehicles: Fig. 3 shows the impact of the number of vehicles on different scheduling approaches with respect to different performance metrics. Increasing the number of vehicles creates additional queries and hence workload in an RSU, resulting in a higher DMR value (Fig. 3(A)). Among all the approaches our proposed CQS approach shows better performance than others with varying workload in the RSUs. MRF shows the poorest performance, because it serves data item considering data item popularity and does not consider unserved query status, hence a query

which tries to access cold data items needs to wait long time for complete satisfaction, which causes some queries to miss their deadlines. FCFS serves the queried data items of a query contiguously. This query-level scheduling contrasts with item-level scheduling in MRF, hence a selected query irrespective of its queried data items status (hot or cold) can go for complete satisfaction quickly. Queries that miss their Deadlines in an RSU comprises both queries that were transferred from neighbor RSUs and those generated locally. Just as DMR increases with increased workload at an RSU, the SRTQ of the system declines with higher workload (Fig. 3(B)). However, as CQS can achieve reasonably lower DMR value than other approaches throughout the changing number of vehicles which also means the CQS approach serves a larger number of queries than other approaches for same workload, both the SRTQ and SER value (Fig. 3(C)) of CQS approach is better than other approaches for the same workload condition. Impact of Query Total Deadline (QTD) value: With increasing upper limit in the value of QTD (according to equation 7) some queries have larger QTD values. A larger QTD helps a query to stay alive longer in the system; it helps a query to be satisfied completely in two ways: (1) Longer QTD value provides a better opportunity to the RSU to transfer that query to more distant RSU having the lowest workload among the neighbor RSUs in a specific direction, (2) an unsatisfied query can have a chance to transfer to multiple RSUs on the route the corresponding vehicle is heading. This is also the reason that with increasing QTD, the DMR of different approaches declines (Fig. 4(A)), SRTR and SER increase (Fig. 4(B) and 4(C) respectively). However, in a standalone approach a query is only feasible until its QLD value expires and doesn’t transfer a query to other RSUs, changing the upper limit of QTD value does not affect QLD of the query in the originated RSU (because dwll ≥ T QLD (RSU )). So, changing the lower limit QTD is Tmax qi i QTD has no effect on the standalone approaches. In the next experiment we change the value of ε from 10 to 5.5 and vary query generation inter-arrival time to keep query arrival in an RSU to a constant rate. Impact of Query Size (Number of data items requested by a query): Fig. 5 shows the impact of query size on the performance of the scheduling approaches. When query size is 1, the performance difference between the cooperative or stand-alone approaches is negligible. However with increasing query size this difference increases significantly. This is because when query size is bigger a query typically consists of a higher number queried data items, so then to satisfy a query query-level scheduling becomes more effective than item-level scheduling i.e. considering unserved query status becomes important. Hence, query-level scheduling approaches such as CQS and FCFS show significantly better performance with increasing query size than an item-level scheduling approach such as MRF in terms of DMR, SRTQ and SER. It can also be seen that CQS performs the best.

Deadline Miss Ratio (DMR)

0.60 0.55 0.50 0.45 0.40 0.35 0.30

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.25 0.20 0.15 0.10 15

30

45

60

0.95

CQS FCFS with CLT MRF with CLT

0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.50 0.45

0.95

Scheduling Efficiency Ratio (SER)

0.65

Success Rate of Transferred Queries (SRTQ)

0.70

0.90 0.85 0.80 0.75 0.70

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.65 0.60

0.40

75

15

30

Number of Vehicles

45

60

0.55

75

15

30

Number of Vehicles

45

60

75

Number of Vehicles

Deadline Miss Ratio (DMR)

0.60

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.55 0.50 0.45 0.40 0.35 0.30 0.25 (X=2, Y=1)

(X=3, Y=2)

(X=4, Y=3)

(X=5, Y=4)

0.95 0.75

0.90

0.70 0.65 0.60 0.55 0.50 0.45

CQS FCFS with CLT MRF with CLT

0.40 0.35

(X=6, Y=5)

(X=2, Y=1)

Query Total Deadline (QTD)

(X=3, Y=2)

(X=4, Y=3)

(X=5, Y=4)

Scheduling Efficiency Ratio (SER)

0.65

Success Rate of Transferred Queries (SRTQ)

Figure 3. Impact of number of vehicles, (A) Deadline Miss Ratio (DMR) of different approaches, (B) Success Rate of Transferred Queries (SRTQ) of proposed CQS approach with different CLT approaches, (C) Scheduling Efficiency Ratio (SER) of different approaches

0.85 0.80 0.75

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.70 0.65 0.60

(X=6, Y=5)

(X=2, Y=1)

Query Total Deadline (QTD)

(X=3, Y=2)

(X=4, Y=3)

(X=5, Y=4)

(X=6, Y=5)

Query Total Deadline (QTD)

Deadline Miss Ratio (DMR)

0.8 0.7 0.6 0.5 0.4

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.3 0.2 0.1 1

1~3

2~4

2~6

4~6

4~8

0.9

1.0

CQS FCFS with CLT MRF with CLT

0.8 0.7 0.6 0.5 0.4 0.3 0.2

Scheduling Efficiency Ratio (SER)

0.9

Success Rate of Transferred Queries (SRTQ)

Figure 4. Impact of Query Total Deadline (QTD) value: (A) Deadline Miss Ratio (DMR) of different approaches, (B) Success Rate of Transferred Queries (SRTQ) of proposed CQS approach with different CLT approaches, (C) Scheduling Efficiency Ratio (SER) of different approaches 0.9 0.8 0.7 0.6 0.5

CQS FCFS with CLT MRF with CLT Standalone FCFS Standalone MRF

0.4 0.3

1

1~3

Query Size

2~4

2~6

Query Size

4~6

4~8

1

1~3

2~4

2~6

4~6

4~8

Query Size

Figure 5. Impact of Query Size: (A) Deadline Miss Ratio (DMR) of different approaches, (B) Success Rate of Transferred Queries (SRTQ) of proposed CQS approach with different CLT approaches, (C) Scheduling Efficiency Ratio (SER) of different approaches

V. C ONCLUSIONS In this work we have explored an VANETs environment where multi-item query with diverse requested data item sizes can be served cooperatively in multi-RSUs environment. We propose a cooperative query serving (CQS) scheduling approach and show that it outperforms different existing scheduling approaches in both a standalone or a cooperative configuration using simulation experiments. R EFERENCES [1] T. Mak, K. Laberteaux, and R. Sengupta, “A multi-channel VANET providing concurrent safety and commercial services,” in Proceedings of the 2nd ACM international workshop on Vehicular ad hoc networks (VANET’05), Cologne, Germany, September 2005, pp. 1–9. [2] O. Maeshima, S. Cai, T. Honda, and H. Urayama, “A roadsideto-vehicle communication system for vehicle safety using dual frequency channels,” in Proceedings of the IEEE Intelligent Transportation Systems Conference (ITSC’07), Seattle, WA, USA, September 2007, pp. 349–354.

[3] Y. Chung and M. Kim, “Effective data placement for wireless broadcast,” Journal of Distributed and Parallel Databases, vol. 9, no. 2, pp. 133–150, March 2001. [4] N. Prabhu and V. Kumar, “Data scheduling for multi-item and transactional requests in on-demand broadcast,” in Proceedings of the 6th international conference on Mobile data management (MDM’05), Ayia Napa, Cyprus, May 2005, pp. 48–56. [5] J. Chen, V. Lee, and K. Liu, “On the performance of real-time multi-item request scheduling in data broadcast environment,” Journal of Systems and Software, vol. 83, no. 8, pp. 1337– 1345, August 2010. [6] K. Liu and V. Lee, “On-demand broadcast for multiple-item requests in a multiple-channel environment,” International Journal of Information Sciences, vol. 180, no. 22, pp. 4336– 4352, November 2010. [7] “Cooperative inter-RSU query transfer.” [Online]. Available: http://www.cs.cityu.edu.hk/∼gnawazali/ CooperativeQueryTransfer.pdf [8] F. Bai, N. Sadagopan, and A. Helmy, “The IMPORTANT framework for analyzing the impact of Mobility on Performance Of RouTing protocols for Adhoc NeTworks,” Journal of Ad Hoc Networks, vol. 1, pp. 383–403, 2003.