Delay Reduction by Providing Location Based Services using Hybrid Cache in peer to peer Networks

Abstract

Now a days, Efficient processing of Broadcast Queries is of critical importance with the ever-increasing deployment and use of mobile technologies. BQs have certain unique characteristics that the traditional spatial query processing in centralized databases does not address. In novel query processing technique, by maintaining high scalability and accuracy, latency is reduced considerably in answering BQs. Novel approach is based on peer-to-peer sharing, which enables us to process queries without delay at a mobile host by using query results cached in its neighboring mobile peers. We design and evaluate cooperative caching techniques to efficiently support data access in ad hoc networks. We first propose two schemes: Cache Data, which caches the data, and Cache Path, which caches the data path. After analyzing the performance of those two schemes, we propose a hybrid approach (Hybrid Cache), which can further improve the performance by taking advantage of Cache Data and Cache Path while avoiding their weaknesses. Cache replacement policies are also studied to further improve the performance. Simulation results show that the proposed schemes can significantly reduce the query delay and message complexity when compared to other caching schemes.

1. Introduction

Broadcast query processing is becoming an integral part of many new mobile applications. Recently, there has been a growing interest in the use of broadcast queries (BQs), which represent a set of queries that retrieve information based on mobile users’ current locations [1,2].

Novel query processing techniques must be devised to handle the following new challenges:

1. Mobile query semantics.

In a mobile environment, a typical BQ is of the form “find the top-three nearest hospitals.” The result of the query depends on the location of its requester.

Caching and sharing of query results must take into consideration the location of the query issuer.

2. High workload.

The database resides in a centralized server, which typically serves a large mobile user community through wireless communication. Consequently, bandwidth constraints and scalability become the two most important design concerns of BQ algorithms [2].

3. Query promptness and accuracy.

Due to user mobility, answer to a BQ will lose their relevancy if there is a long delay in query processing or in communication. For example, answers to the query “find the top-three nearest hospitals” received after 5 minutes of high-speed driving will become meaningless. Instead, a prompt, albeit approximate, top-three nearest hospitals, may serve the user much better. This is an importance issue, as a long latency in a high workload wireless environment is not unusual.

The wireless environment and the communication constraints play an important role in determining the strategy for processing BQs. In the simplest approach, a user establishes a point-to-point communication with the server so that the user queries can be answered on demand. However, this approach suffers from several drawbacks. First, it may not scale to very large user populations. Second, to communicate with the server, a client must most likely use a fee-based cellular-type network to achieve a reasonable operating range. Third, users must reveal their current location and send it to the server, which may be undesirable for privacy reasons [12]. A more advanced solution is the wireless broadcast model [1, 9, 18]. It can support an almost-unlimited number of mobile hosts (MHs) over a large geographical area with a single transmitter. With the broadcast model, MHs do not submit queries. Instead, they tune in to the broadcast channel for information that they desire. Hence, the user’s location is not revealed. One of the limitations of the broadcast model is that it restricts data access to be sequential. Queries can only be fulfilled after all the required on-air data arrives. This is why in some cases, a 5-minute delay to the query “find the top-three nearest hospitals” would not be unusual.

Alleviating this limitation, we propose a scalable low-latency approach for processing BQs in broadcast environments. Our approach leverages ad hoc networks to share information among mobile clients in a peer-to -peer (P2P) manner [10, 11].

The rationale behind our approach is based on the following observations:

 

2. Wireless Data Broadcast

In general, there are two approaches for mobile data access. One is the on-demand access model, and the other is the wireless broadcast model. For the on-demand access model, point-to-point connections are established between the server and the mobile clients, and the server processes queries that the clients submit on demand. For the wireless broadcast model, the server repeatedly broadcasts all the information in wireless channels, and the clients are responsible for filtering the information. The advantage of the broadcast model over the ondemand model is that it is a scalable approach. However, the broadcast model has large latency, as clients have to wait for the next broadcast cycle broadcasting cycle. If a client misses the packets that it needs, it has to wait for the next broadcast cycle.

To facilitate information retrieval on wireless broadcast channels, the server usually transmits an index structure, along with data objects. A well-known broadcast index structure is the (1, m) [9] indexing allocation method. As we can see in Fig. 1, the whole index is broadcast preceding every 1/m fraction of the data file. Because the index is available m times in one cycle, it allows a mobile client easy access to the index so that it can predict the arrival time of its desired data in a timely manner, and once it knows the arrival time, it only needs to tune into the broadcast channel when the data bucket arrives. This mechanism is important for battery-based devices.

Fig. 1.The Hilbert-curve-based index structure. The numbers represent index values.

Thus, the general access protocol for retrieving data on a wireless broadcast channel involves three main steps [15]:

However, nearly all the existing spatial access methods are designed for databases with random access disks. These existing techniques cannot be used effectively in a wireless broadcast environment, where only sequential data access is supported. Zheng et al. [19] proposed indexing the spatial data on the server by a space-filling curve. The Hilbert curve was chosen for this purpose because of its superior locality. The index values of the data packets represent the order in which these data packets are broadcast. For example, the Hilbert curve in Fig. 1 tries grouping data of close values so that they can be accessed within a short interval when they are broadcast sequentially. The MHs use on-air search algorithms [19] to answer LBSQs (kNN and WQs) over data that arrives in the order prescribed by the Hilbert curve.

2.1 Spatial Queries

Existing work focus on two common types of spatial queries, namely, kNNqueries and WQs [17]. With R-tree-based [15] spatial indices, depth-first search (DFS) and best first search (BFS)[8] have been the prevalent branch-and-bound techniques for processing NN queries. The DFS method recursively expands the index nodes for searching NN candidates. At each newly visited nonleaf node, DFS computes the ordering metrics for all its child nodes and applies pruning strategies to remove unnecessary branches.

When a leaf node is reached, the data objects are retrieved, and the NN candidates are updated. Comparatively, the BFS technique utilizes a priority queue to store nodes to be explored through the search process.

The nodes in the queue are sorted according to their minimum distance (MINDIST) to the query point. During the search process, BFS repeatedly dequeues the top entry in the queue and enqueues its child nodes with their MINDIST into the queue.

When a data entry is dequeued, it is inserted into the result set. For WQs that find objects within a specified area, the R-tree families[3,16] provide efficient access to disk-based databases. Basically, an R-tree structure groups objects close to each other into a minimum bounding rectangle (MBR), and a range query only visits MBRs that overlap with the query area.

2.2 Cooperative Caching

Caching is a key technique to improve data retrieval performance in widely distributed environments Hara and Madria proposed three data replica allocation methods in ad hoc networks by considering the access frequency from MHs to each data item and the status of the network connection With the increasing deployment of new P2P wireless communication technologies, P2P cooperative caching becomes an effective sharing alternative With this technique [4, 7]. MHs communicate with neighboring peers in an ad hoc manner for information sharing instead of relying solely on the communication between remote information sources.

P2P cooperative caching can bring about several distinctive benefits to a mobile system: improved access latency, reduced server workload, and alleviated point-to- point channel congestion. In this research, we leverage the P2P caching technique to alleviate the inherent access latency limitation in wireless broadcast environments.

2.3 Sharing-Based NN Queries

Fig. 2 shows an example of an on-air kNN query based on the Hilbert curve index structure. At first, by scanning the on-air index, the k-nearest object to the query point is found, and a minimal circle centered at q and containing all those k objects is constructed. The MBR of that circle, enclosing at least k objects, serves as the search range. Consequently, q has to receive the data packets that covers the MBR from the broadcast channel for retrieving its knearest objects.

Fig. 2.An on-air kNN query example. The numbers represent index values

As shown in Fig. 2, the related packets span a long segment in the index sequence, that is, between 5 and 58, which will require a long retrieval time. The other problem with this search algorithm is that the indexing information has to be replicated in the broadcast cycle to enable twice scanning. The first scan is for deciding the kNN search range, and the second scan is for retrieving k objects based on the search range [19].

Therefore, the Sharing-Based Nearest Neighbor (SBNN) query approach to improve the preceding on-air kNN query algorithm. The SBNN algorithm attempts to verify the validity of k objects by processing results obtained from several peers.

 

3. Overview

The wireless data broadcast model has good scalability for supporting an almost-unlimited number of clients [9]. Its main limitation lies in its sequential data access: the access latency becomes longer as the number of data items increases. If we can provide (approximate) answers to spatial queries before the arrival of related data packets, we will overcome the limitation of the broadcast model.

A novel component in the methodology is a verification algorithm that verifies whether a data item from neighboring peers is part of the solution set to a spatial query. Even if the verified results constitute only part of the solution set, in which case the query client needs to wait for the required data packets to get the remaining answers, the partial answer can be utilized by many applications that do not need exact solutions but require a short response time (for example, the query “What are the top-three nearest hospitals?” issued by a motorist on a highway).

3.1 Nearest Neighbor Verification

When an MH q executes SBNN, it first broadcasts a request to all its single-hop peers for their cached spatial data. Each peer that receives the request returns the verified region MBR and the cached POIs to q. Then, q combines the verified regions of all the replying peers, each bounded by its MBR, into a merged verified region MVR (the polygon in Fig. 3). The merging process is carried out by the MapOverlay algorithm [6] (line 4 in Algorithm 1). The core of SBNN is the NN verification (NNV) method, whose objective is to verify whether a POI oi obtained from peers is a valid (that is, the top-k) NN of the MH q.

Fig. 3.Because e1 has the shortest distance to q and ||q,o1||<=||q,e1||,POI is verified as a valid NN of MHq.

Let IP denote the data collected by q from j peers pi,… pj. Consequently, the merged verified region MVR can be represented as

MVR =p1.VR U p2.VR U ….. U pj VR

Suppose that the boundary of MVR consists of k edges, IE ={e1,e2, . . . ek}, and there are l POIs, ={o1, o2,…..ol} inside the MVR. Let es ϵ IE be the edge that has the shortest distance to q. An example is given in Fig. 3, where k=10, and e1 has the shortest distance to q.

The NNV method uses a heap H to maintain the entries of verified and unverified POIs. Initially, H is empty. The NNV method inserts POIs to H as it verifies objects from MHs in the vicinity of q. The heap H maintains the POIs in ascending order in terms of their euclidean distances to q. nverified objects are kept in H only if the number of verified objects is lower than what was requested by the query. The NNV method is formalized in Algorithm 1. Since the verified-region merging process dominates the algorithm complexity, the NNV method can be computed in O(n log n+i log n) time, where n is the total edge number of the two merged polygons, and i is the number of intersection points.

If k elements in H are all verified by NNV, the kNN query is fulfilled. There will be cases when the NNVmethod cannot fulfill a kNN query. Hence, a set that contains unverified elements is returned. If the response time is critical, a user may agree to accept a kNN data set with unverified elements, where the objects are not guaranteed to be the top kNNs. However, the correctness of these approximate results can be estimated and will be discussed in the next section. If the result quality is the most important concern, the client has to wait until it receives all the required data packets from the broadcast channel. Nevertheless, the partial results in H can be used to decrease the required data packets and thus speed up the on-air data collection.

3.2 Approximate Nearest Neighbor

We calculate the probability that the unverified ith NN o of a query point q is actually the true ith NN of q. The reason that o cannot be verified is because there is a region that is not covered by q’s neighboring peers. As long as a POI exists in the region, o cannot be q’s ith NN. We denote such a region as o’s unverified region. We assume that the POIs are Poisson distributed in our environment based on our experiments of several common POI types (gas stations, grocery stores, etc.) with chi-square tests [13, 14]. The probability of finding another POI in the unverified region Ui of an unverified POI oi can be calculated with respect to the area of Ui. We formulate the correctness of an unverified POI based on the probability model

 

4. Proposed Basic Cooperative Cache Schemes

In this section, we propose two basic cooperative cache schemes and analyze their performance.

4.1 System Model

Fig. 4.1 shows part of an ad hoc network. Some nodes in the ad hoc network may have wireless interfaces to connect to the wireless infrastructure such as wireless LAN or cellular networks. Suppose node N11 is a data source (center), which contains a database of n items d1; d2; . . . ; dn. Note that N11 may be a node connecting to the wired network which has the database. In ad hoc networks, a data request is forwarded hop-by hop until it reaches the data center and then the data center sends the requested data back. Various routing algorithms have been designed to route messages in ad hoc networks. To reduce the bandwidth consumption and the query delay, the number of hops between the data center and the requester should be as small as possible. Although routing protocols can be used to achieve this goal, there is a limitation on how much they can achieve. In the following, we propose two basic cooperative caching schemes: Cache Data and Cache Path.

Fig. 4.System Model

4.1.1 Cache the Data (CacheData)

In CacheData, the node caches a passing-by data item di locally when it finds that di is popular, i.e., there were many requests for di, or it has enough free cache space. For example, in Fig. 4.1, both N6 and N7 request di through N5, N5 knows that di is popular and caches it locally. Future requests by N3, N4, or N5 can be served by N5 directly. Since CacheData needs extra space to save the data, it should be used prudently. Suppose the data center receives several requests for di forwarded by N3. Nodes along the path N3 _N4 _ N5 may all think that di is a popular item and should be cached. However, it wastes a large amount of cache space if three of them all cache di.

To avoid this, a conservative rule should be followed: A node does not cache the data if all requests for the data are from the same node. As in the previous example, all requests received by N5 are from N4, which in turn are from N3. With the new rule, N4 and N5 do not cache di. If the requests received by N3 are from different nodes such as N1 and N2, N3 will cache the data. If the requests all come from N1, N3 will not cache the data, but N1 will cache it. Certainly, if N5 receives requests for di from N6 and N7 later, it may also cache di. Note that di is at least cached at the requesting node, which can use it to serve the next query.

This conservative rule is designed to reduce the cache Space requirement. In some situations, e.g., when the cache size is very large or for some particular data that are interested by most nodes, the conservative rule may decrease the cache performance because data are not cached at every intermediate node. However, in mobile networks, nodes usually have limited cache spaces and we do not assume that some data are interested by all nodes. Therefore, the conservative rule is adopted in this paper.

4.1.2 Cache the Data Path (CachePath)

The idea of CachePath can be explained by using Fig. 4.1 Suppose node N1 has requested a data item di from N11. When N3 forwards the data di back to N1, N3 knows that N1 has a copy of di. Later, if N2 requests di, N3 knows that the data center N11 is three hops away whereas N1 is only one hop away. Thus, N3 forwards the request to N1 instead of N4. Note that many routing algorithms (such as AODV[and DSR provide the hop count information between the source and destination. By caching the data path for each data item, bandwidth and the query delay can be reduced since the data can be obtained through fewer number of hops. However, recording the map between data items and caching nodes increases routing overhead.

In the following, we propose some optimization techniques. When saving the path information, a node need not save all the node information along the path. Instead, it can save only the destination node information, as the path from current router to the destination can be found by the underlying routing algorithm. In CachePath, a node does not need to record the path information of all passing-by data.

For example, when di flows from N11 to destination node N1 along the path N5 _ N4 _ N3, N4 and N5 need not cache the path information of di since N4 and N5 are closer to the data center than the caching node N1. Thus, a node only needs to record the data path when it is closer to the caching node than the data center. Due to mobility, the node which caches the data may move. The cached data may be replaced due to the cache size limitation. As a result, the node which modified the route should reroute the request to the original data center after it finds out the problem. Thus, the cached path may not be reliable and using it may adversely increase the overhead. To deal with this issue, a node Ni caches the data path only when the caching node, say Nj, is very close. system tuning threshold, called TH, when CachePath is used.

 

5. Hybrid Caching Scheme

The performance analysis showed that Cache Path and Cache Data can significantly improve the system performance. We also found that Cache Path performs better in some situations such as small cache size or low data update rate, while CacheData performs better in other situations.

5.1 Hybrid Algorithm

To further improve the performance, we propose a hybrid scheme to take advantage of Cache Data and Cache Path while avoiding their weaknesses. Specifically, when a node forwards a data item, it caches the data or path based on some criteria. These criteria include the data item size si, the TTL time TTLi, and the Hsave. For a data item di, the following heuristics are used to decide whether to cache data or path

These threshold values should be set carefully as they may affect the system performance. Their effects and how to set them are studied through simulations

5.1 Hybrid Cache Algorithm

Section 5.1 shows the algorithm that applies these heuristics in Hybrid Cache. In our design, caching a data path only needs to save a node id in the cache. This overhead is very small. Therefore, in Hybrid Cache, when a data item di needs to be cached using Cache Data, the path for di is also cached. Later, if the cache replacement algorithm decides to remove di, it removes the cached data while keeping the path for di. From some point of view, Cache Data degrades to Cache-Path for di. Similarly, Cache Path can be upgraded to CacheData again when di passes by.

The parameter TTLi and Ts play an Important role in the final performance ,because TTLi (TTLi –Time to Live for data item i ) decides when to forward the request to the data center for requested data if it is not available in the nearest neighbor.

Ts is Threshold level for client cache ,if threshold level exceeds in client cache ,the client can forward its request to nearest neighbor and cache replacement policies also done.

 

6. Performance Evaluation

The performance evaluation includes three sections. The simulation model is given in Section 6.1. In Section 6.2, simulated environments and screenshots are shown. Section 6.3 shows the performance graph of cache data, cache path and hybrid cache from the analytical results.

6.1 The Simulation Model

The simulation is based on ns-2 with Cygwin Environment.. In our simulation, AODV (Ad Hoc On Demand Distance Vector) is used as an underlying routing algorithm. Since it is better than DSDV and I-DSDV(Improved Direct sequence Distance Vector). The aforesaid fact is justified by following results.

Packet Delivery Fraction of AODV

From the Figure 6.1 a) it is shown that AODV performs better when the number of nodes increases. This is because nodes that become more stationary will lead to more stable path from source to destination. DSDV performance drops as number of nodes increases, because more packets are dropped due to link breaks. I-DSDV is better than DSDV especially when the number of nodes is between 20 and 35. I-DSDV improves the PDF as it finds a new route to the destination when link breaks exist.

Fig. 6.1 a).Packet Delivery Fraction

End to End Delay of AODV

From the Fig 6.1 b) it is clear that AODV didn’t produce so much delay even when the number of nodes increased. It is better than the other two protocols. The performance of IDSDV is slightly better than DSDV especially when the number of nodes is between 15 and 30. It shows that, the I-DSDV protocol improved the DSDV but it was slightly lower than AODV when the nodes are higher. We assumed that the wireless bandwidth is 2 Mb/s, and the radio range is 200m to 250m.

Fig. 6.1 b).End to End Delay

In Fig 6.1b) when we increase the number of nodes initially, the delay gets increased for AODV, DSDV and I-DSDV. Later stage AODV and I-DSDV reach concave point ( indicates delay reduction) because AODV supports many nodes .When we Compare DSDV with AODV(Fig 6.1b), AODV produces low delay even when nodes are higher in count.

Two concave spots are observed at 10 and 25 for blue and yellow line. And one concave spot is observed at 10 for purple line. Because at 10, AODV, DSDV and I-DSDV’s delay are reduced at 25 when increasing the number of nodes (only AODV and IDSDV delay get reduced) .Finally the AODV performed better in delay reduction and Packet delivery (Fig 6.1 a) and was used as an underlying routing algorithm for our approach.

In paper [8] R tree based spatial data structure is used with k nearest neighbor algorithm for effectively finding nearest mobile node but not to deal with cache path and forward the requests (our proposed algorithm supports). In paper [15] Branch and Bound R -tree traversal algorithm is used to find the nearest neighbor object to a point and not supporting more number of pages accessed (our proposed algorithm supports). In paper [17] Knn is given as a input to SBNN and SBWQ (sharing based window query) which is effective only in caching but the data does not deal with cache path.(our proposed algorithm supports).

The experiment on both real and synthetic data sets are analyzed and corresponding results are plotted (Fig 6.1 c) From the results we can understand that the number of pages accessed is high in Hybrid cache algorithm even when increasing the number of neighbors (120).

Fig. 6.1 c).Page Access Comparison between Hybrid Cache, R-Tree and Branch and Bound

In Paper [4] COCA (co-operative caching) leads to flood cache, however, as our method uses cache path, flood cache can be avoided.

In Paper [7] discussion is about COCA with Push based information .In this Server sends repeatedly data to client. This also leads to fill the cache of the client.

Experimental results (Fig 6.1 d) shows the comparison between hybrid cache, COCA and COCA with push based information ; we can see the delay percentage which is reduced for hybrid cache.

Fig. 6.1 d).Delay Comparison between Hybrid Cache ,COCA and COCA with Push based

6.2 Simulation

Experiments were run using different workloads and system settings. The design of wireless scenario is shown in the Fig 6.2 (a). In this caching of node and caching of path is shown. In Fig 6.2(b) and Fig 6.2(c) the system model of the ad hoc network is shown. In this, seven nodes are shown. In that, one node acts as a Data Center and the remaining nodes have peer to peer connection.

Fig. 6.2 a).Wireless scenario

Fig. 6.2 b).Ad Hoc Model

Fig. 6.2 c).Request of node 1 and Response from Data center (node 0)

As per the Hybrid cache algorithm Cache Path and Cache Data have been done with the help of ns2 simulator. Caching of data is done when the node 0 requests the data from the data center. If the request is again from the node 0, the node 0 itself analyzes for the data and gets the results .If some other node in this ad hoc network requests for the same data again, a node which has the path id of the node1 will respond to the request.

6.3 Performance Graph

The Fig. 6.3 shows the comparison graph of cache data, cache path and hybrid cache. Comparison has been done by taking transmission Range as a parameter. From the Analytical results we can understand that hybrid cache methodology gives gradual growth while increasing transmission range. Hybrid cache is better than cache data and cache path. In this scenario we have considered cache data and cache path. We have considered cache data whenever it is essential (depending upon threshold level and size of a data) and also considered cache path (depending upon threshold of Time to live of a data item).

Fig. 6.3.Transmission Range Experiments

 

7. Conclusion

This paper presented a novel approach for reducing the spatial query access latency by leveraging results from nearby peers by applying Hybrid cache Mechanism. Specifically, we proposed three schemes: Cache Path, Cache Data, and Hybrid Cache. In Cache Data, intermediate nodes cache the data to serve future requests instead of fetching data from the data center. In CachePath, mobile nodes cache the data path and use it to redirect future requests to the nearby node which has the data instead of the faraway data center. Hybrid Cache takes advantage of CacheData and CachePath while avoiding their weaknesses. Cache Replacement policies are also studied to further improve the cache performance. Simulation results showed that the proposed schemes can significantly reduce the query delay when compared to SimpleCache and significantly reduce the message complexity when compared to FloodCache The experiment results indicate that our method can reduce the access to the wireless broadcast channel by a significant amount, for example, up to 90 percent.