A Scalable and Robust QoS Architecture for WiFi P2P Networks

Transcription

1 A Scalable and Robust QoS Architecture for WiFi P2P Networks Sathish Rajasekhar, Ibrahim Khalil, and Zahir Tari School of Computer Science and Information Technology, RMIT University, Melbourne (sathish, ibrahimk, zahirt)@cs.rmit.edu.au Abstract. Peer-to-Peer (P2P) resource sharing between mobile devices in Wireless Fidelity (WiFi) hot-spots environment is a challenging problem. This would require an infrastructure with automated process for registering new mobile devices, as well as authentication and authorisation of existing devices. Further, issues such as maintenance, and updating the state information, as devices join and leave the P2P network; optimising route selection and protection of the existing mobile devices from malicious devices are crucial. To address these issues, we propose a generalised architecture and a dynamic protocol for effective and optimal file transfer between devices. We use quality of service (QoS) capacity-tohop count ratio, routing algorithm, to find an optimal mobile device for a service request. The goal and contribution of this paper is to provide a scalable, robust and reliable architecture incorporating QoS; effective and optimal communication for P2P networks in a cooperative manner. 1 Introduction Internet is exponentially exploding and as changes to the electronic technology evolve, P2P will play a pivotal role in enduring information sharing, resource management enabling interoperability and QoS. 80% of data collected on an edge router at France Telecom IP backbone network was P2P traffic [2]. Also on the Sprint IP backbone P2P and unknown traffic type was about 80% [6]. As the processing power and memory of mobile devices such as personal digital assistant (PDA) is increasing, resource sharing between these devices becomes more rampant in the near future and users will start demanding QoS. Unfortunately, current P2P technology does not provide a stable, scalable, reliable and robust architecture addressing QoS issues. We propose a super-peer based architecture for P2P resource sharing in the core network addressing signalling, reliability, scalability and QoS issues. A super-peer is a static and powerful device, which intelligently and collectively manages the entire operation of file transfer and assists peers in finding optimal QoS path. Most peer devices exhibits different characteristics with respect to their capabilities such as bandwidth, storage, and processing power. These capabilities are exploited by super-peers. Peer devices update their location and state information to super-peers. Each device has the same range to send and receive information. These devices form a connected multi-hop wireless network. R.K. Ghosh and H. Mohanty (Eds.): ICDCIT 200, LNCS 337, pp. 65 7, 200. c Springer-Verlag Berlin Heidelberg 200

2 66 S. Rajasekhar, I. Khalil, and Z. Tari Cluster 1 Cluster 2 SP SP WAP Access Point Routers SP 3 Core Network SP Cluster 3 Cluster Fig. 1. Mobile Devices in WiFi Hot-spots For example, let us consider an airport where passengers with hand held devices wants to share resources. Super-peers SP1, SP2, SP3 and SP as shown in Figure 1 are connected to the core network through the wireless access point (WAP). If a passenger within SP1 wants to download a file, the passenger s device is sensed by the WAP router. SP1 initiates a file search in the network. If the file is found in devices that belong to SP2 and SP3, an optimal QoS path based on capacity-to-hop count [12] is selected. Most of the existing work done in P2P systems are to effectively look for data [3][15]. Napster [1], Gnutella [1] and Freenet [1] are P2P technologies developed and have problems such as single point of failure, signalling and scalability issues. A. Klemm et al in [11] propose a protocol called Optimised Routing Independent Overlay Network (ORION). This protocol works on a centralised approach leading to a central point of failure. In [] the authors define XREP protocol, but repositories on peers cause load burden. None of these existing approaches address scalability, reliability, load sharing and QoS path selection criterion. Our goal is to provide a generalised scalable, reliable and robust architecture for resource sharing incorporating QoS path selection for P2P systems. We propose a Mobile Authentication and Resource exchange (MARX) protocol for this purpose. Resource sharing is done using the state information obtained by IDMaps [5]. IDMaps provides the network distances in terms of available bandwidth between two hosts. The available bandwidth information obtained by IDMaps is used by capacity-to-hop count ratio routing [12] algorithm to help super-peers find optimal QoS paths. We also incorporate caching schemes for efficient information retrieval and replicate super-peer database, to encounter single point failure. Our objective is to propose a QoS architecture and

3 A Scalable and Robust QoS Architecture for WiFi P2P Networks 67 investigate such issues. We also show that use of super-peer based QoS architecture helps in authentication and authorisation of devices. This paper is organised as follows. Section 2 is related work and puts our work in context. The architecture is detailed in section 3. In section, the protocol and routing issues are discussed with an example to illustrate capacity-to-hop count routing algorithm. We conclude in section 5 with directions for future work. 2 Related Work A huge body of work has been done in efficient information retrieval systems such as CAN [13], Chord [15], Pastry [1], and Tapestry [9]. These information retrieval approaches implement distributed hash tables (DHT). A query is associated with a key and the query is routed to a peer holding the key value in DHT to locate information. Napster [1], Gnutella [1], Freenet [1] etc are some of the early P2P approaches. Napster, a centralised scenario, leads to single point failure. In Gnutella, a decentralised model, peers generate redundant signals all over the network. Freenet tries to overcomes these issues to some extent, but do not address QoS issues. A protocol for dynamic file sharing was proposed by Klemm et al [11], called ORION. This protocol has redundant data at nodes and is prone to single point failure. A cluster based architecture was proposed in [10], where peers are clustered based on content awareness. This defeats the purpose of a mobile environment as peers should be allowed to move freely from one WiFi hot-spot to another. Also, the authors indicate that the quality of the clusters formed is not guaranteed. Damiani et al in [] propose a secure and robust reputation mechanism for choosing reliable resources using XREP protocol. They emphasise on two repositories, a resource repository and a servant repository on every peer device. In a WiFi environment, holding two repositories by a mobile device is a daunting task and is not achievable. Our approach attempts to eliminate the above pitfalls by providing a a scalable, robust, reliable architecture and protocol ensuring QoS path selection amongst peer devices in WiFi hot spots. 3 WiFi P2P File Sharing Architecture: An Overview Mobile devices are capable of exchanging information using Wireless Fidelity. WiFi defines the wireless technology in the IEEE specification including the wireless protocols a, b, and g in a WiFi hot-spot. A WiFi hot-spot is a location in which wireless technology exists and is available for use to consumers [8]. We propose to build a generalised architecture with existing technologies. The main components of our proposed architecture are detailed in Figure 2 and discussed below. Super-Peer: A new wave in P2P systems, providing advancements of having a centralised database in a decentralised environment. They are responsible for

4 68 S. Rajasekhar, I. Khalil, and Z. Tari Applications Interface Query Info Module Valid Query Update Info Module Stats. Super Cache F Cache L Cache QoS Broker IDMaps Peer CAN Pastry Lookup Module CHORD Tapestry Act. Devices Rgd. Devices LISSP SLA Profile Fig. 2. Internal Architecture Overview servicing a request from a peer. Super-peers are powerful static devices that helps in selecting optimal QoS paths. Also, super-peers help in registration, authorisation, authentication and accounting of a device. Query Information Module: The functions are two fold. Module Valid checks for peer authorisation. This is done by checking the super-peer database. The module Query initiates search to find suitable resources that satisfies the QoS requirements of the generated query. Update Information Module: The availability of the optimal peer device for file transfer based on the QoS requirements is updated. The module Stats updates the cache with the most popular files accessed based on the frequency of down-load. Caching: A popular file may be accessed more frequently. Instead of searching the same file every time, these files are cached. We classify cache as file cache (F-Cache) and location cache (L-Cache). F-Cache stores the most frequently accessed files. L-Cache stores the most frequently accessed location. Caching reduces signalling and overhead traffic. QoS Module: This module consists of QoS interface and IDMaps Interface. QoS Broker ensures that the QoS requirements between the service requester and the device that is offering service is met. It establishing a reliable connection in an optimal manner using capacity to hop count ratio routing algorithm [12]. QoS broker thus updates the information regarding the file and its statistics to update info module. Internet Distance Map Service (IDMaps) a global architecture for Internet host distance estimation and distribution [5], provides quickly and efficiently the network distances in terms of metrics such as latency or band-

5 A Scalable and Robust QoS Architecture for WiFi P2P Networks 69 width between Internet hosts. Higher level services collect distance information and bandwidth to build a virtual distance map of the Internet and estimate the distance between any pair of IP addresses. For example in Figure 1, a device in SP1 requests for a resource that is found in SP2 and SP3. IDMaps provides the available bandwidth information between SP1 and SP2, and SP1 and SP3 respectively. Based on the information provided, a routing decision is made by the QoS broker using capacity to hop count routing algorithm. Thus IDMaps provides network distance in terms of latency or bandwidth and it is scalable. Lookup Module: A search for requested resource is carried on using any one of the existing lookup modules such as CAN, Pastry, Chord, Tapestry algorithms. LISSP: The Local Information Source Super Peer, is a centralised information database. This aids in authentication of a mobile device. The services for any peer device is processed though LISSP. It contains a list of active devices Act devices and registered devices Rgd devices. Act. devices are devices currently present in the super-peer cluster. Rgd. devices are devices registered in the super-peer but currently not available. SLA is the service level agreement between WiFi customers and WiFi service providers. The SLA profile is maintained in the LISSP module. A device requesting service is called service receiver, and that offering the service is called as service provider. The LISSP checks the device s profile for authorisation. If the device is registered in the WAP database or the LISSP, then file request query is generated and serviced. The F-Cache and L-Cache is checked for the resource. The update info module informs the service receiver regarding the requested file or its location. Else, the lookup module initiates search for the requested resource. IDMaps gathers available bandwidth of service providers and informs the QoS broker. The broker selects an optimal device based on capacity to hop count ratio algorithm, which is discussed in the routing sub-section. In the next section, we discuss the properties and working of the proposed protocol. Mobile Authentication and Resource exchange (MARX) Protocol The proposed dynamic MARX protocol aids in le discovery, connection setup and maintenance, le transfer and termination of connection. The protocol takes into consideration the search initiated by the lookup module through IDMaps and selects the optimal peer based on capacity-to-hop count routing algorithm..1 MARX Overview and Assumptions For example, from Figure 1 a mobile device in super-peer SP1 wants to download a file. The steps involved are summarised as follows: 1. The device is registered and authenticated with a unique address under SP1 or some other super-peer.

6 70 S. Rajasekhar, I. Khalil, and Z. Tari 2. MAC address of a new device is made known to the local super-peer. 3. The device requests its super-peer (SP1) for a file.. The device gets a response from SP1 regarding the file availability; after initiating a search, QoS broker identifies the service provider, which may be within SP1 or a different super-peer. 5. QoS Broker selects the optimal peer based on capacity-to-hop count ratio. 6. The device requests for connection from the identified service provider. 7. The device receives data from the service provider peer. 8. The device disconnects from service provider after file transfer. Some basic assumptions are made to define our protocol. The communicating peer devices may be in the same or different cluster of the same super-peer or in a different super-peer altogether. The clusters are independent and may be under one super-peer. Sharable information from peer devices are periodically updated to their super-peer LISSP database. A cluster of peer devices connected to the super-peer through a WAP is depicted in Figure 3. The peer devices are sensed through the WAP and its database for authentication as they enter the cluster. SP WAP WAP WAP 1 WAP Database 2 LISSP Database 3 OTHER SUPERPEERS Cluster Cluster Cluster 6 5 Fig. 3. Logical Overview of Clusters and Super-Peer Connectivity Fig.. Global Device Authentication During the process of authentication, one of the three possibilities rise. The device information can be found in the clusters WAP or LISSP database of the super-peer; if not found, a search is initiated amongst other super-peers as shown in Figure. Steps 1 and 2 of Figure fail to authenticate the peer device and hence the other super-peers are queried for authentication as shown in step 3. The steps, 5 and 6 informs the device regarding its authenticity; thirdly, the device may be a new device accessing the WAP. The new device has to register first with a super-peer. Once a query is generated by a device, a search is initiated in the super-peer cache, within the super-peer or a device in different super-peers. Resource in Cache: Figure 5 illustrates how communication between the device and cache takes place. Once a query is initiated (step 1), the cache is checked for the requested resource (step 2). Steps 3 and confirm that the resource is

7 A Scalable and Robust QoS Architecture for WiFi P2P Networks 71 present in cache. The device (service requester) requests for the file (step 5) and establishes a connection. Step 6 represents file transfer between the cache and the service receiver. Step 7 indicates disconnection. LISSP database 1 Query Cache 1 LISSP SPn Fig. 5. Cache Access Fig. 6. Cluster Access Resource Within Cluster: If the cache does not contain the requested file, query is passed on to the lookup module. The lookup module initiates a search within its LISSP database (step 1) and the super-peer (step 2) as shown in Figure 6. If one or more devices within the cluster has the file (step 3), then the QoS broker identifies the optimal peer, based on certain policies, such as distance, load, how long the device was in the cluster and informs the service receiver. Load is defined as the number of devices that are using the services of the device identified. Based on the QoS broker information, update info module informs the service receiver, regarding service provider (step ). Steps 5, 6 and 7 of Figure 6 represents file transfer request, the file transfer and termination of connection amongst peer devices within a cluster. Resource in different Super Peer: The service provider at times cannot be found in the same vicinity or in the same super-peer. Also, peers may not respond due to other activities. Hence, a request is issued to other super-peers and networks as shown in Figure 7. The QoS broker gets an updated list of super-peers that are willing to share information through IDMaps. The QoS broker decides the optimal super-peer based on certain parameters such as delay, available bandwidth, and hops in terms of capacity to hop count ratio [12]. The information is then passed on to update info module and data transfer takes place. Step 1 in Figure 7 initiates a query. Step 2 issues a search. Step 3i return multiple hits. The QoS broker selects optimal service provider device and informs the service requester through update info module as in step of Figure 7. Steps 5, 6 and 7 indicate file transfer and its completion.

8 72 S. Rajasekhar, I. Khalil, and Z. Tari Device Visit P/N Fshare UId Super Peer Database MAD 1 MAD 2 SP 3 SP 2 N P 1 n1 1 n SP 1 SP 1 1 LISSP 2 Provider F Cache L Cache MAD 1 MAD 2.. SLA MAD N SP P 1 n3 FF SP 3 MAD N 6 3i 5 Mac Id Dyn IP Other WiFi Cloud 7 Fig. 7. Access to Different Networks Fig. 8. LISSP Database and Connectivity LISSP Replication: LISSP database prohibits unauthorised users from gaining access to the network resources. Figure 8 depicts a LISSP database, SLAs between different clusters and the mobile access devices (MAD). Each MAD holds information in the LISSP database regarding its identity (MAC address), super-peers visited, and list of sharable information. Devices update the LISSP as and when they have new information to share. Also the devices inform LISSP where they were first registered. Any centralised system is prone to single point of failure. Hence we propose replication of the LISSP database in super-peers. Each super-peer node will have its nearest two neighbouring super-peer LISSP databases as backups thus eliminating single point failure..2 Routing QoS path selection is the process of selecting a path based on QoS requirements such as bandwidth or delay [7]. Our QoS routing algorithm maximises the available path capacity to hop count as described in [12]. The objective of QoS routing is to eliminate the inaccurate state information, the proposed QoS routing can achieve higher efficiency and better optimisation. As we increase the number of hops, the ratio path capacity to hop count is applied. Mathematically, C avl (P ) h(p ) The available capacity is denoted as C avl, P a path and h(p ) the hop count of a path P. The available capacity to hop count ratio is computed as follows: (1) C avl (P )=min l P (b avl (l)) (2) The routing algorithm uses the available capacity to hop count ratio as the criterion for optimality, and its correctness is proved in [12]. This algorithm obtains paths with maximal ratio of available capacity to hop count from the source super-peer node to all the other super-peer nodes. We illustrate QoS

9 A Scalable and Robust QoS Architecture for WiFi P2P Networks SP1 10 SP0 60 SP3 SP2 30 Fig. 9. Routing Illustration using Capacity-to-Hop Ratio Algorithm path selection using an example as shown in Figure 9. If super-peer SP0 wants to route to super-peer SP3, the optimal route has to be determined. Initially bandwidth (B) at the super-peer SP0 is set to and all the other super-peers, SP1, SP2 and SP3 to zero. Predecessor of all super-peers are set to NIL. After the first hop from SP0, we have B[SP1]=50, B[SP2]= and B[SP3]=0. We also have the hop count h[sp1] = h[sp2] = h[sp3]=1. After the second hop from SP0, there is no change for SP1, while B[SP2]=25 and h[sp2]=2. The details for SP3 are successively updated to B[SP3]=5 and h[sp3]=2, followed by B[SP3]=10 and h[sp3]=3. The complexity of the algorithm is similar to that of Bellman-Ford algorithm. 5 Conclusion In this paper, we have described a scalable, robust, reliable, generalised QoS architecture for P2P systems. We provide an optimal QoS path selection based on capacity-to-hop count ratio routing algorithm for efficient file transfer between devices in a P2P network. We propose a dynamic protocol MARX, to effectively carry out file transfer using customer-provider SLAs. Our super-peer based architecture do not allow unauthorised devices to access the network and protects existing devices from malicious devices. A new device is registered and the whole process is automated without any human intervention. While we have not carried out simulation on P2P networks to show the effectiveness of our proposed architecture, the QoS path selection algorithm applied in our current approach was validated in IP networks [12]. In our future work, we would like to further simulate/emulate our QoS architecture and QoS path selection algorithms for P2P networks. We envisage a prototype model for validation and verification purposes. Also, we would like to investigate and propose new techniques for load balancing among peer devices in a P2P network.

10 7 S. Rajasekhar, I. Khalil, and Z. Tari Acknowledgement This work is proudly supported by the Australian Research Council (ARC), under discovery Scheme, DP References 1. K. Abeer and M. Hauswirth. Peer-to-Peer information systems: concepts and models, state-of-the-art, and future systems, Proceedings of ACM SIGSOFT Software Engineering Notes, Vol. 26, Issue 5, pp , N. B. Azzouna and F. Guillemin, Experimental analysis of the impact of peerto-peer applications on traffic in commercial IP networks, [Online]. Available: 3. H. Balakrishnan, M. F. Kaashoek, D. Karger, R. Morris and I. Stoica, Looking Up Data in P2P Systems Proceedings of Communications of the ACM, Vol. 6, No. 2, pp. 3-8, E. Damiani, S. De. C. Vimercati, S. Paraboschi, P. Samarati and F. Violante, A Reputation-Based Approach for choosing Reliable Resources in Peer-to-Peer Networks, Proceedings of ACM Conference on Computer and Communications Security, pp , P. Francis, S. Jamin, C. Jin, D. Raz, Y. Shavitt, L. Zhang, IDMAPS: A Global Internet Host Distance Estimation Service, IEEE/ACM Transactions on Networking, C. Fraleigh et al, Packet-Level Traffic Measurements from the Sprint IP Backbone, IEEE Network, 17(6), pp. 6-16, R. Guerin and A. Orda, Computing Shortest Paths for Any number of Hops, IEEE/ACM Transactions on Networking, 10(5), pp , C. Hesselmen, H. Eertink, I. Widya and E. Huizer, A Mobility-aware Broadcasting Infrastructure for a Wireless Internet with Hot-spots, Proceedings of WMASH, pp , K. Hildrum, J. Kubiatowicz, S. Rao and B. Zhao, Distributed Object Location in a Dynamic Network, In Proceedings of 1th ACM Symp. on Parallel Algorithms and architectures, I. A. Klampanos and J. M. Jose, An Architecture for Information Retrieval over Semi-Collaborating Peer-to-Peer Networks, Proceedings of ACM Symposium on Applied Computing, pp , A. Klemm, C. Lindemann, and O. P. Waldhorst, A Special-Purpose Peer-to-Peer File Sharing system for Mobile Ad Hoc Networks. [Online] 12. S. Rajasekhar, B. Lloyd-Smith, and Z. Tari, QoS Path Routing based on Capacity to Link Ratio in Networks, Proceedings of International Conference on Networks, Parallel and Distributed Processing, and Applications, Japan, pp , S. Ratnaswamy, P. Francis, M. Handley, R. Karp and S. Shenker, A Scalable content-addressable network, In Proceedings of ACM SIGCOMM, A. Rowstron and P. Druschel, Pastry: Scalable, distributed object location and routing for large-scale peer-to-peer systems, In Proceedings of the 18th IFIP/ACM International Conference on Distributed Systems Platforms, I. Stoica, R. Morris, D. Karger, M. F. Kaashoek and H. Balakrishnan, Chord: A Scalable Peer-to-Peer Lookup Service for Internet Applications, Proceedings of SIGCOMM, pp , 2001.