P2P is supposed to revolutionize the way one perceives traditional client/server computing. Here's a lowdown on this technology. by A. K. Vanwasi

P2P (Peer-to-Peer) is the new buzzword that is supposed to change the way one perceives traditional computing.

A P2P network enables its 'member nodes' to search for and share content and spare computing resources in real-time, on the Internet. P2P information sharing paradigm is expected to enhance workgroup productivity and support community activity.

The P2P philosophy differs significantly from conventional client/server computing. Currently, most of the Internet applications use client/server computing. Here, a central server stores information and distributes it to clients in response to their requests. However, the information stored is static and can be updated only by the central server.

P2P networking is an alternative to client/server computing. It is server-less computing. There exists many varieties of P2P networks like Gnutella, Grid computing, SETI@home, etc. Member nodes of a particular P2P network share common networking software and may be located on the Internet or any other private network. They actively participate in information search and distribution. Each member node also contributes to information repository.

P2P networks also enable sharing of spare computing power and storage of data among member nodes. Thus, P2P network enables a web of millions of computers to act as a gigantic distributed virtual super-computer.

Here, we provide an overview of various P2P networks and related issues.

P2P basics
A better way to understand P2P computing would be to grasp the related components and terminology.

P2P network: A P2P network comprises of online 'member nodes' located on the Internet. Each member node has common P2P networking protocol software and an individual information repository. A member node can search for and directly share information repository and computational resources of other member nodes.

Member Node: A member node can act both as a client and server. It also maintains IP addresses of few other member nodes.

Open P2P network: Such a P2P network is able to distribute all type of content i.e. music, video and data files.

Content-based search: A P2P network supports content-based search. A member node i.e. source, queries the P2P network for a specific content. The P2P network software on other member nodes matches the information repository for the requested content. If a match is found on a member node, then the source is a sent response. Now, the source can directly download the content.

Dynamic indexing: A member node can only be accessed if it is online. This ensures that the information repository is always updated.

Server-based indexing: A single server maintains the index of contents of all the member nodes. It replies to request of a member node for content with list of IP addresses of other member nodes having the requested content.

P2P multicast: A source node sends only one copy to a multicast group. This copy is replicated along downstream, taking advantage of the overlapping distribution paths. This eases traffic on the Internet.

P2P network load-balancing: A P2P network load-balancing technique monitors traffic and the demand profile for a particular information item and then redistributes content to ease the load on an individual node. The load-balancing policy may also locate the content close to usage.

P2P networks
A P2P network can be something as simple as a LAN workgroup, or something on the lines of Napster the popular file-sharing program that lets you share files over the Internet. Here are few examples of existing P2P networks, and their functionality.

LAN workgroups: Small P2P networks have their origin in LAN environments. Workgroups in a LAN use popular operating systems like Apple Macintosh or Windows for sharing files.

Napster: Napster is a popular online music file sharing global network. Here, member nodes are located on the Internet. Every member node has proprietary Napster client software loaded on the computer.

Napster uses a central server to maintain a directory of music files that are available with different member nodes. Due to use of a central server, Napster is not a true P2P network.

Its operation involves:

• Request by a client member node for a MP3 song to central server.
• On receipt of a request, the server searches for the desired music file and returns the list of member nodes having the requested content.
• Now, client member establishes connection with the member node having the desired file and downloads the music file.

Gnutella: It is an open, online P2P network. It does not require a central server for indexing data files and is the least proprietary.

Gnutella protocol is a specialized software and communication protocol. It enables a host computer to function as both a client and server. To use it, a software application called 'Servent' (Server + Client) is left running on each member node.

Using the Gnutella protocol, a member node can search for other online member nodes. Usually, each member node keeps track of four or five other member nodes. Thus, there is a web of virtual interlinked member nodes.

Gnutella protocol enables a member node to designate a file as a shared file. To coordinate communication among member nodes, the current Gnutella protocol (Version 0.4) defines five descriptors: Ping, Pong, Query, Query-Hit and Push.

Steps involved in communica-tion are:

• To find other online member nodes, a member node sends a Ping packet that announces the presence of a member node.
• When another member node receives a Ping packet, it responds by sending a Pong packet comprising IP address, port number and directory of shared files. Further, the receiving node forwards the copy of Ping packet to other member nodes.
• Gnutella Query packet enables a requesting member node to search for the desired data files on other member nodes, on the basis of content.
• If a member node has found a match, then it responds to the requesting member node by sending Query-Hit packet comprising IP address of member node, port number and data file name.
• On receipt of Query-Hit, requesting member node can download the desired data file. Normally, every Query packet has 'Time To Live' field that defines the lifetime of Query packet in the network. This avoids indefinite circulation of Query packet in the network.

Freenet: It is a Java-based open P2P network. Like Gnutella, it also does not use a central server. It operates across platforms. It uses different routing and caching schemes. This enables efficient file distribution and scalability.

Morpheus and Kaza: These are two popular server-less video file swapping software programs. Member nodes with these software can swap video files. They are also used to transport songs and grainy rerun of TV shows.

Distributed Computing
Distributed computing is a popular application area for P2P networks. It is based on the premise that during normal operation, 90 percent of desktop computer's processing cycle remains unused. An estimate suggests that Internet connected PCs offer at least an aggregate 10 billion MHz of processing power and 1000 TB of storage. P2P networking enables organizations to exploit these globally distributed spare computing resources. Further, the distributed computing decentralizes computation tasks and reduces project cost and administration. Here are some key applications for distributed computing.

SETI @ home: This P2P network uses thousands of online PC nodes located on the Internet to help in search of extraterrestrial intelligence. Users participate by running a free program that downloads and analyzes radio telescope data. There are rare chances that a computer will detect any faint signal of an extraterrestrial civilization but this has not prevented enthusiasts from participating in the SETI program.

Grid Computing: A Grid links PCs to form a virtual super computer using 'Globus' middleware. Globus is downloadable from the Internet (www.globus.org). It enables organizations to create an infrastructure that allows them to share data, applications, processing power and storage. It provides an automatic way of discovering resources and scheduling applications to run computation tasks on shared computational resources available on the Internet.

There are now hundreds of grids composed of millions of PCs known as nodes that are linked all around the globe. Grid computing should enable corporates to have cheap but massive computing power on tap and more importantly on a pay-as-needed basis.

It is estimated that in the near future, scientific and commercial computing grids will be linked together into national grids. Eventually, national grids will be linked into a global grid.

Drug Research Program: As philanthropy, Intel has initiated a drug research program. To participate in this program, one can download an application from www.intel.com/cure.

To find a potential cure or new drug, researchers must evaluate the cancer-fighting potential of hundreds of millions of molecules. It is estimated that the task will require a minimum of 24 million hours of number crunching.

Commercial Application: Data synapse's 'Web Proc' product enables a customer to tap unused processing power of PCs to conduct complex financial and business computations. The company pays people to use their idle computers.

'Web Proc' software breaks complex computing jobs into smaller tasks and using the Internet distributes them in encrypted form to PCs having spare computing capacity.

Endeavor Technology has developed P2P products that enable an enterprise to sell digital goods and information directly from their own computers without a website or server.

Peer issues

Like most new technologies, P2P is still evolving. There are some factors that affect the smooth functioning of P2P networks. Solution to these problems will be key for the success of P2P.

• Bandwidth: Contrary to asymmetrical channel requirement of client/server computing, P2P networks require symmetrical communication channel i.e. same upstream and downstream channel bandwidths. Further, P2P computing applications are bandwidth hungry and calls for better quality of service (QoS). Telecom service providers have to make suitable provision for it.
• Computing power: PCs and computers used in P2P applications will require more computing power to manage the communication and server overhead.
• Security: In P2P networking many applications can access other PC's hard disk directly. Here, security is the main concern. Downloading of files from other computers makes the system vulnerable to viruses. Further, it is required for the computers to authenticate other machines.
• Interoperability: Member nodes of a P2P network have variety of operating systems, networking technologies and other platforms in business applications. Thus, advanced interoperability techniques will be necessary. XML-based interoperability technologies hold lot of potential for P2P computing.
• Copyright: P2P networks allow members to share information that may, subsequently, raise issues of copyright infringement, intellectual piracy and undesirable spread of malicious contents. Thus, P2P computing may form basis of many chaotic and lawless communities. We only can expect that as P2P communities evolve, they become more stable.
• QoS: Success of P2P computing requires better quality of service and contents. User feedback-based performance score can provide information for improving connection reliability and content quality.
• IP addressing: A P2P network is a transient community. They do not have always-on connectivity to the Internet. In other words, member nodes do not have permanent IP addresses. This calls for a new IP addressing scheme.

Conclusion
P2P computing offers a better alternative to client/server computing for searching and sharing contents. However, success of P2P computing requires satisfactory resolution of security issues, upgradation of PCs and provisioning of high-capacity symmetrical bandwidth network.

P2P is still in an evolutionary phase to have a real impact on client/server computing in the near term. However, it has the potential to revolutionize the way we perceive traditional computing, and in the long term offer significant cost benefits.

A. K. Vanwasi is GM (R&D),ITI LTD. Naini. He can be reached at vanwasi_nni@itiltd.co.in