Client-side proxies | Master's thesis, May 2000 | |
<< Previous [ Conclusions ] Next >> |
Table of contents | |
Before drawing any conclusions from the previous sections, let us recapitulate the aim and purpose of this thesis. The overall context is the Internet and its abundance of resources. The purpose here is to investigate the merits of the client-side proxy approach as a way to help users find interesting information through content processing, adaptation and information retrieval. From a general viewpoint, this section focuses on the questions posed in the introduction: When are client-side proxies better and when could other approaches be preferable? Do existing client-side proxies realise the potential benefits of the approach? Are there ways to improve them? Table 1 provides parts of the answers, summarising the characteristics of different approaches for content processing.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Table 1. Characteristics of different solutions. | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
The first thing to consider is whether to use client-side proxies at all. Compared to other solutions, is there anything that gives a proxy the upper hand?
One of the trademarks of proxies is that they have direct access to the content stream. Sitting in the middle of communication, they can easily intercept everything of interest. This is clearly an advantage compared to client plug-ins and remote Web services. A plug-in is subject to the good will of its parent environment. It might get complete access to the content through the client, but inherits the limitations of an integrated application. Web services have only indirect access to communication between client and server. An example is the Anonymizer, described in section 4.1. A user must manually request documents on the Anonymizer site or through a special text-field in an anonymised document. If the user does not deliver this information, the Anonymizer has no access to the content. Was it a proxy, it would automatically intercept all requests without placing cognitive demands on the user. Hence, the proxy approach is more exhaustive than Web services. It is also more exhaustive in the sense that it can handle virtually any kind of communication - Web documents, email, ftp, news, telnet, etc. Integrated clients also have this ability and direct access to the content, but they are less exhaustive. An integrated application can not process content accessed through other client applications. Since plug-ins rely on their parent applications for content access, the same limitations apply to them.
Because the proxy in theory can intercept anything from anywhere, it has big potential to perform privacy-enhancing processing, such as encryption and anonymisation. That it can apply this processing before the content leaves the local machine is a strength it has in common with stand-alone clients. Again, these applications can only handle their own communication, while the proxy can process all communication before releasing it to the network. When Web services are involved, the initial communication is always unprotected.
In general, developers should consider the client-side proxy approach when the task at hand demands direct access to the content stream. If the task also involves privacy protection and exhaustive interception of different kinds of communication from different client applications, the proxy solution is clearly the best choice. The Freedom proxy of section 4.1 is a good example.
Since a client-side proxy is located on the end-user's local machine, it has access to the full functionality and processing power of the local operating environment. Like integrated clients and plug-ins, but unlike Web services, it can perform demanding tasks close to the destination of the content, where it is most efficient. An illustrative example is PureSight, described in section 4.5, that uses demanding artificial intelligence algorithms for content analysis. Even if Web services run on powerful servers, a large number of users share the processing power. It is easier to provide a fast and reliable service if the functionality is reasonably simple.
Although it is better to perform more advanced and power-demanding processing locally, dedicated servers are better at relatively simple but large-scale operations. Web page indexing performed by search engines and online directories is one example where the local environment is simply too small to store the information. An interesting solution would be to use the processing power of client-side proxies in conjunction with the power of large-scale central servers to facilitate collaborative processing. A real-world example of this is the SETI@home project, part of the Search for Extraterrestrial Intelligence program at UC Berkeley [SETI 99]. Interested Internet users download a small part of the massive amounts of data collected through the SETI programs, and when their local computer has processed the data, the results are returned over the Internet. The project does not use client-side proxies, but it shows the strength of collaborative processing. A client-side proxy could download or gather data, process it and deliver the results to a central server, utilising the local processing power.
The processing power available to an application has obvious impact on what it can do and what levels of sophistication it can reach. Somewhat simplified, a local approach has potential to be more sophisticated than a remote Web service. If the task involves applying demanding algorithms to relatively small amounts of data, such as single Web documents, the local approach is preferable. It does not matter whether it is a proxy, an integrated client or a plug-in, as long as they have access to the local machine. In this sense, the local approach can be more sophisticated. On the other hand, if the involved algorithms are simpler, but the processed data more extensive, a high-end Web services could be better. The massive storage capacity required is better utilised if many users share the resource.
Since the Web came into being, the major usability focus has been on ease of learning for novice users. Simplicity has been the obvious gain, but at the cost of sophistication, especially for experienced users. As an example, the Anonymizer Web service is simple and straightforward, but decidedly not as sophisticated as the client-side proxy approach of Freedom. Although Web services often are simpler than local applications, the functionality does not have to be trivial [Nielsen 00]. As users become more loyal to specific web-sites and as they come to depend on web-based functionality in their daily work, these services must adapt to the needs of experienced users. Improving navigation by providing something similar to keyboard shortcuts in local applications is one example. As Web services evolve towards more sophisticated functionality, they will challenge the client-side applications. A sign of the times is the propaganda for thin clients and application service providers.
Sophistication can also be realised by aggregating the efforts of several actors. Most proxies, both client-side and others, support aggregation of behaviour under user control. Non-proxy plug-ins and applications might also support some notion of aggregation, but not in the simple and standardised way of the proxy. Proxies commonly support aggregation through chaining, or more rarely, through extensibility. Both these approaches are described in section 5.2.3. Through its support for aggregation, the proxy approach is more flexible and allows individual users to extend the functionality by simply adding another proxy or proxy extension.
All the examined proxies have shown, to different degrees, that one of the major benefits of the proxy approach is client independence. By placing the processing functionality in a layer independent of client brands and versions, the proxy can deliver the same functionality regardless of user preferences. The proxy has this property in common with Web services.
On the other hand, an integrated application has a closer relationship with the user that it can exploit for detailed monitoring of user behaviour, down to single mouse movements and keyboard actions. While it is true that a client-side proxy can analyse the outer aspects of user behaviour, such as what resources are requested and time between requests, a more integrated solution can create more fine-grained and advanced user profiles.
Connected to integration is the question of transparency. An approach is fully transparent if it works completely in the background or appears as an integrated part of the functionality of another application. A stand-alone client is obviously not transparent, and neither are Web services. A plug-in is transparent, since it is an extension to the functionality of the parent application and appears to be a part of this application. A client-side proxy can also be fully transparent. It can run completely in the background, as the content blocking applications of section 4.5, or it can integrate both interaction and presentation with the client application like the WebMate proxy or the Blueberry framework (sections 5.1.2 and 6, respectively). These approaches integrate closely with the content, appearing to be part of the requested documents or the overall functionality of the client application.
Although some of the proxies examined in this work are independent of platform as well as client, it is not an inherent quality. Proxies rely on the same technologies available to stand-alone applications and plug-ins. If platform independence is crucial, a web-based service is the natural choice as the only one providing true independence. Otherwise, platform-specific applications have many benefits, whether they are proxies or not. Integration with a familiar environment can make the application more visually attractive and easy to use, as opposed to the sluggish interfaces exhibited by the independent proxies examined in this work. Another important benefit is that direct use of platform-specific functionality can improve the overall performance of the application.
Platform independence could be important for software developers. That proxies use standardised network protocols and network functionality common to many operating systems indicates that this approach could be more platform-independent than applications with closer platform integration. When an application is developed for multiple platforms, using independent solutions can reduce the time, cost and complexity of the development process. An application developed as a platform-independent proxy is widely available for testing, with the option to tailor subsequent production releases to the most important target platforms to provide the benefits of platform-specific applications.
For the end-user, platform independence is probably not an important issue. Client-side proxies are supposed to run on single machines, and single machines usually provide a single operating platform. Of course, some users work on multiple platforms, and if they want to use the proxy functionality on every machine, a platform-independent solution is preferred. In general, users are presumably more interested in the gains of platform-dependence - easy installation, a familiar user interface, better performance, etc - than in the vision of platform independence.
Even if client-side proxies have several virtues, as we turn to overall performance and usability we must acknowledge that other solutions generally are better. Although content processing always has negative impact on performance, unless the processing is explicitly aimed at enhancing performance, proxies can be even worse than the other approaches. A major reason is the local socket communication required for most proxies, while integrated clients and plug-ins work in the client application environment. Close platform integration, as exhibited by for example the Freedom and PureSight proxies, could be a way to alleviate the performance impact, since low-level communication methods are more efficient.
Close integration could also improve usability, which is a weakness with many proxies. In general, they are more difficult to install and configure since they require configuration of both the proxy itself and the client applications whose content they want to process. Platform-specific approaches could alleviate this burden. If ease of use is crucial, a web-based approach should also be considered. As already discussed, a major feature of these services is high usability, and they require no installation at all.
The ease with which an application can be uninstalled is also a usability factor, but proxies are generally as easy or difficult to uninstall as other solutions. Most proxies use the same uninstallation procedures available to all kinds of applications. However, it is an issue when proxy chains are involved. If a proxy that is part of a chain is uninstalled, the chain is broken and the user must reconfigure the neighbour proxy. When this happens, uninstalling a client-side proxy is more complicated than uninstalling an ordinary application.
It might seem discrediting to platform-independent techniques that the independent proxies of this examination exhibits slower, clumsier and visually unattractive interfaces compared to the platform-specific systems. The lack of common design guidelines for platform-independent applications is partly to blame, but such guidelines will probably evolve as the approach matures.
Apart from purely technical considerations, there might be situations where it is not possible or desirable to use client-side proxies, due to legal or ethical considerations. The problem with ad removers such as WebWasher has already been mentioned in section 4.3. If many users decide to remove advertisements, it could become harder for providers to supply free services. Anonymisers such as Crowds or Freedom might also be viewed unfavourably. Companies could forbid their employees to use them, since they make it difficult for administrators to monitor users' online behaviour. Shopping sites could refuse to accept orders from anonymised connections, since anonymisation makes it harder to trace fraudulent users. Furthermore, it could be irritating for a information provider if the information is processed and changed on the way to the user. All factors combined, there is a risk that online actors will take steps to inhibit the use of such applications, unless they handle these issues in a manner that is acceptable to all parties. Of course, these concerns apply to any content processing application, but many of the existing client-side proxies focus on this kind of tasks.
What is most distinguishing of the client-side proxy approach is the natural and direct access to the content stream. Several potential benefits that can make the client-side proxy better at content processing originate in this closeness. Let us take a closer look at how the proxies of today utilise the potential of the approach, and how to make them better in the future.
The first potential benefit is, obviously, to easily access, analyse and adapt the content. In the context of this work, this is what client-side proxies are all about. The close tie to the network could also make retrieval of additional information a natural part of proxy functionality. Apart from BackLink in section 6.4 and the SELECT proxy for collaborative rating (section 4.2) that provides information about other users' ratings, this approach is not so common. Through analysis of the communication flow between client and servers, a client-side proxy could also build models of the user, that could be used to refine the behaviour of the proxy to provide support optimised for individual users. WebMate is the only example of this approach. In general, functionality for building user models and for retrieving additional information is scarce among the proxies examined in this work. This is an area with great potential, and it might be good to take further advantage of it.
Close to the content but separated from the content presentation, the proxy approach is basically client-independent. Although some proxies, for example WebMate and SELECT, use techniques that slightly circumscribe this independence, it is still a strong point in favour of the proxy approach. A user can take advantage of the functionality of a client-independent proxy regardless of the application used for presentation, and this is definitely something that existing and future proxies should uphold.
Related to independence is transparency. With functionality placed in a layer separate from presentation, client-side proxies can do their work in the background in the same manner as operating systems services. If we leave out installation and configuration, most proxies perform in the background. However, the focus on content processing poses a problem. It is common that processing generates information that should be visible to the user. The standard solution is that the proxy provides an application environment of its own, making it less transparent. An alternative is to provide this information as part of the content, when the content protocol allows this. In reality, this kind of integration is feasible only with HTML content. Blueberry carries this notion to the extreme, incorporating the complete user interface in the processed content. WebMate is more cautious - the interface is accessible through the content but displayed in its own windows. That incorporation of content and interface is possible is also an effect of the direct access to the content. It is assumed here that processing creates interesting results, and that these results should be displayed as close to the working environment as possible. However, this is an opportunity that should be used with caution, since it imposes great structural changes on the requested document and occupies a largish part of the client application's workspace.
The potential for sophistication, either by utilising the local machine for demanding processing or by aggregation of functionality, is partly fulfilled by the client-side proxies of today. Most processing is relatively simple text matching and filtering, but there are also attempts at more powerful processing, most notably in WebMate and PureSight. Compared to the simpler approaches, the content analysis performed by PureSight minimises the need for user interaction and manual updates, resulting in a more usable application. This is an example from which others could learn. Although most proxies perform relatively simple processing, they generally support sophistication through aggregation. The common way to combine the functionality of several proxies is chaining. This is straightforward and rather flexible, but a well executed extensible approach might be better for performance and usability, conducting all processing and user interaction within a single application. The extensible proxies Muffin and ByProxy (section 5.2.4) partly live up to this notion, but the higher-level content abstraction and integrated interface of Blueberry shows a possible way to utilise the potential even more.
Compared to other approaches, client-side proxies have architectural strength. The combination of direct access to the content, client independence, access to the local operating environment and the inherent support for aggregate behaviour is a compelling argument to use proxies for sophisticated content processing. The problem is that even if the architecture has merits, other solutions generally exhibit better usability and performance. Most of the examined proxies will probably be considered only by advanced users, while the other will prefer the simpler and more familiar alternatives - integrated clients, plug-ins and Web services. There are exceptions, such as WebWasher and PureSight, that combine the strengths of the proxy with the usability of integrated applications, but this is not the usual case.
There is also the risk that client-side proxies will not be used, simply because they are not discovered by potential users. Most of the proxies examined here have a low profile, at least compared to heavily advertised integrated clients and Web services. It is more rule than exception that marketing strength is more important than technical merits in deciding which solution will be commonly accepted and used.
On the good side, there is a close and natural tie between proxy, content and network. As use of the Internet increases and client-side machines and applications become more integrated with networks and remotely hosted services, applications with network capabilities have the competitive edge. This is clearly an advantage for the client-side proxy, since networking is the foundation of its existence. If this advantage is utilised together with a stronger focus on usability issues, the future of the client-side proxy might not be so bleak.
The primary focus of this thesis has been the potential merits of client-side proxies for content processing, but several related areas should also receive attention. The legal aspects of content processing and adaptation are interesting. Changing content provided by others might be a copyright violation, and displaying retrieved Web pages within a framework such as Blueberry might be viewed unfavourable by the information providers. A survey of the opinions of these providers regarding client-side proxies and content processing could be of value. If integration of user interface in processed documents is better than clean separation, and how it should be done to be accepted by users also deserves a more in-depth answer. As wireless communication becomes more important, proxies could be a bridge between earthbound and ethereal resources. Whether client-side proxies have anything to contribute to these mobile environments could be investigated. Finally, the specific processing tasks involved need to be continually examined and improved. There has been much research regarding these topics, such as methods to retrieve information fulfilling the needs of individual users, building sophisticated user profiles, making navigation easier, etc. However, due to the fast pace of technology-changes and growth of available information, this area requires constant attention.
<< Previous [ Conclusions ] Next >> |
Table of contents | |