Graffiti Server — Design and Implementation

While data on file systems and data on the Web have traditionally been organized in a hierarchical structure, tagging has emerged as a viable technology for dealing with large collections of data. Tagging involves attaching descriptive keywords to data objects such as files and URLs. Most current implementations of tagging restrict the scope of tags to the website or application in which they were created. We have designed and implemented the Graffiti system to explore the collaborative use of tags across applications, computers and users. The Graffiti system is made up of two key components. The first is a client application which the user utilizes to manage the tags on their local file system. The second is a server application that enables collaborative metadata management and sharing. The Graffiti server constitutes a back-end database and a server application which provides Graffiti clients with access to shared metadata. This document describes the design and implementation of the current version of the Graffiti server. It includes a complete description of the current installation of the server as well as detailed instructions for extending the capabilities of the server. 1 Project Introduction Locating data that resides on file systems has traditionally been very different than finding web-pages. Locating data on the local file system is closely tied to the act of “filing” data. In contrast, locating data on the Web is more closely related to “finding”. The disparity between locating data on local systems and the Web can be tied to fundamental differences in the way the user interacts with the system. Recently however, tagging has emerged as a new model for locating data on the Web and aspects of it may be applicable to local and shared file systems. To determine how tagging techniques might be applied to file systems we have create a tool called Graffiti which adds tagging capabilities to existing systems and allows us to collect usage data about how users utilize tagging in file systems. On file systems, users usually store files in a hierarchical structure with the hope that this careful placement will make latter retrieval easier. For example a user may place a new document they are working in a directory named, \home\myhome\documents\project1. When it comes time to retrieve this document the user can rely on the structure of their folder hierarchy to know that within their home folder they probably stored their documents in the folder documents and if the document pertains to project1 they have a good idea as to where they placed their file. In this “filing” scenario the user spends time to carefully place their data in a location that they can easily deduce later. In contrast to the “filing” model used on file systems, locating data on the Web can best be described as “finding”. In this model, since the user is not responsible for placing the data, they must deduce the data’s full location based on what they know about its content. In a simple usage model users utilize search engines to find websites and lists of favorite URLs to easily return to the site at a later time. Traditional search engines on the Web, as well as in recent file systems, concentrate on automatic indexing. The main challenge of this approach is to rank matching results to a query. On the web, Google’s PageRank [7]is addressing this by taking the link structure of the web into account. Based on anecdotal evidence, this search technique worked so well that sets of a few keywords became shorthand for URLs and greatly diminished the value of maintaining personal bookmarks. Recently a new model, tagging, has emerged for finding data on the Web. Tagging, in the general case, consists of attaching descriptive text to objects. Many applications of tags have been focused on helping users locate data. Websites such as Flickr [2] and del.ico.us [1] have demonstrated that tagging can effectively replace hierarchal organization schemes and can efficiently organize large collections of data. Using similar, methods applications such as Apple Computer’s iPhoto [3] have included tagging capabilities to help manage collections of data on the local computer. One common drawback that these schemes have is that they do not extend beyond the scope of a single application or website. Tags can be applied to many problems besides data location. For example tags can be used to identify files that are to be included in an action. In this scenario