Tuesday, April 17, 2007

Challenges in realizing mobile P2P networks

This is a short paper I wrote sometime ago about the challenges in realizing scalable mobile p2p networks.

Mobile p2p: A list of challenges and todos

Mobile peer-to-peer networks that share content among mobile devices using multiple network interfaces have been proposed recently. We highlight the important research issues in this area in order to foster further discussion in the important multidisciplinary problems surrounding mobile peer-to-peer networks.

1 Introduction and Motivation

Newer mobile devices such as cell phones and PDAs are capable of connecting to heterogeneous networks over multiple interfaces such as WLAN, WiMAX, cellular 3G networks (UMTS, GPRS, etc.), USB (to PCs), and Blue tooth. Simultaneously, generous amounts of persistent storage such as flash memory and mini-hard drives make these devices candidates for in-network storage and dissemination of multimedia content in a mobile p2p network. Such p2p collaboration can result in significant cost savings: for example, by limiting the amount of data downloaded over expensive cellular data links and instead downloading data over WLAN or Blue tooth from a nearby mobile device, there can be significant cost savings for mobile device users.

There are a number of technical challenges in mobile p2p system design. These can be broadly categorized into content discovery, resource allocation (shared wireless channel contention, bandwidth, battery power), security (content integrity, user anonymity), high churn (due to mobility and unreliable wireless channels), and building incentives for users to participate in mobile p2p networks. The objective of this paper is to enumerate these challenges in order to spur further research in these areas. This list of challenges will also serve as a checklist for mobile p2p system design.

Before we digress into the specific research problems in mobile p2p networks, it is insightful to visualize the concept of a mobile p2p network. Figure 1 shows various mobile devices in a region covered by a cellular data network. These devices are also equipped with WLAN and/or Blue tooth (near-network) interfaces over which ad-hoc networks are formed in proximity situations, such as when many users carrying mobile devices are riding a public bus. These devices hold content that may be interesting to other riders on the bus, for example, a user may have downloaded a newscast into his device earlier that day.

Near-networks are highly dynamic and change rapidly as users carrying the mobile devices move and toggle their device interfaces on or off. The persistence of the various interface connections on mobile devices also varies, with near-network interfaces being more prone to channel interruptions than cellular data connections. Similarly, the costs and throughputs of the various network interfaces vary significantly.

2 Challenges in mobile p2p

There are several important open issues in mobile p2p systems that we enlist below.

1. Content discovery Near-networks change quickly and unpredictably due to mobility and network interfaces on devices being turned on and off. A distributed hash table content discovery algorithm may not function in this environment, especially since the content available in a mobile p2p network will itself change quickly with device mobility. A reactive (on-demand) flooding content discovery approach may be feasible for a small p2p network. Another approach may be to use a centralized directory server accessed through the Internet where mobile devices register their content at the time of joining and leaving the p2p network (assuming Internet connectivity).

2. Session Mobility Session mobility is one of the primary issues of mobile device networks. Mobile devices should seamlessly and transparently handover p2p sessions between interfaces. For example, when a user moves out of a WLAN access point range, the p2p application should switch over to the cellular data network interface, locate the content on this new network, and recommence the p2p session. The handover should be content based, not device based, in the sense that the system should strive to reconnect to the relevant content, and not necessarily the same peer as before.

3. Power efficient protocol design Conserving power is of prime importance, given that mobile devices usually run on batteries. Power efficient design has to be a part of the mobile p2p protocol itself. For example, a centralized content discovery lookup will avoid flooding the p2p network with content discovery query packets, thus conserving power on the mobile devices comprising the network.

4. Delay tolerant protocol Content dissemination in mobile p2p networks will be highly asynchronous due to the transient nature of the underlying physical network. Moreover, a p2p session may span long periods of disconnected operation. The p2p protocol should adapt to long periods of network disconnects.

5. Human mobility Modelling the spread of information in mobile p2p networks is closely related to human mobility. For example, a content file is ‘transferred’ from the near-network of a bus to a cafe’s when a user moves from the bus to the cafe. Designing around human mobility models is crucial for the proper functioning of mobile p2p protocols.

6. Broadcast property of wireless networks Transports such as WLAN offer the possibility of a one-to-many broadcast in the p2p network. This is useful when one device shares popular content with it’s (nearby) peers. The challenge here is exploiting the broadcast property without incurring excessive power drain on only one broadcasting peer. Moreover, long periods of broadcast create channel contention with other competing sessions on the same transport channel.

7. Costs and incentives In addition to bandwidth, battery consumption, and storage space is also a cost to users. Modeling these in incentive mechanisms will significantly influence user participation in mobile p2p networks.

8. Content integrity Mobile p2p networks are arguably more prone to content integrity problems than their wired counterparts because a malicious peer may join a mobile p2p network, pollute p2p content, and leave undetected, owing to the inherent mobility in these systems. Moreover, in a purely near-network p2p setting, downloading checksums of data from a centralized location (as is done, for example, in Bittorrent) is not possible.

9. User privacy Traditional wireline p2p networks provide a degree of privacy by only revealing the IP address of the other peers in the p2p network (only traceable up to the ISP). A near-network mobile p2p network may reveal the identity of a user, especially in a small area. For example, people in a bus downloading content off another mobile device know that the content is coming from a mobile device belonging to someone amongst them.

10. IP rights management Suitable DRM mechanisms for (Internet) disconnected near-networks will need to be developed for any mobile p2p network that serves legal content. Mobile p2p networks may exist independent of Internet connectivity, and this makes download-time centralized licensing impossible. One possible solution to this problem is prepaid licence ‘credit’ that users purchase in anticipation of future licensing requirements when disconnected from the Internet.

3 Concluding remarks

Mobile p2p networks have a different set of design constraints than their wired counterparts. More importantly, users will probably use mobile p2p networks for applications other than just traditional p2p file sharing. Therefore the design of these systems needs to be carefully thought of, rather than just trying to port legacy p2p applications to mobile devices. The challenges listed in this paper need to be addressed suitably while designing these new p2p networks of the future.

All rights reserved. Any reproduction or copying only with consent of author.

No comments: