The next version of the Windows operating system (codenamed "Longhorn") has a new storage subsystem (codenamed "WinFS"). In this article, we will try to understand the need for "WinFS"; define "WinFS", its type system, and its data model; and learn how to write applications that take advantage of "WinFS".

Why Does Windows Need a New Storage Subsystem?
The hardware industry is confidently striding towards conquering the 3T challenge (a teraflops processor, a terabyte hard disk, and a terabits per sec bandwidth). Increase in the size of hard disk storage has been complemented by the exponential increase in the production of digital data. The amount of digital data being born daily is so phenomenal that the pen, the paper, and the typewriter have achieved the status of endangered stationery. Digital data is stored by operating systems on magnetic media - like hard disks. Windows uses file systems like NTFS and FAT32 to organize data on hard disks that are divided into logical drives. Each drive has a root folder, which can have one or more files and folders. Data is stored in files.

Data has personality. It can be documents (text, DOC, PDF, RTF, PostScript formats), images (bitmap, GIF, JPEG, PNG, WMF formats), audio (WAV, MP3, WMA, AIFF, formats), video (MOV, MPEG, AVI, ASF, QuickTime formats), and more. Along with data, these formats also store rich metadata. NTFS and FAT32 are capable of reading only basic metadata from files, like their name, size, date modified, and type (derived from the file extension). So, if I want to search all songs of a particular artist, I can't do it just by using the file system. I'll need a specialized application like Windows Media Player, which is capable of reading the rich metadata provided by all audio files.

What is "WinFS"?
"WinFS" is the Windows storage platform of the future. It solves many of the problems associated with current file systems by storing data and its metadata together. This way data can be organized, searched, shared, and related depending upon what it is and not what it's called (filename)! "WinFS" also makes it easy to deal with non-file-backed data like personal contacts, e-mail messages, and event calendar.

"WinFS" allows a user to organize data with great flexibility. Using "WinFS", a user can group data according to common characteristics, create a customized containment hierarchy to hold data, and associate one piece of data to another via relationships. This organizational flexibility provides the user with greater flexibility when searching for data. The user can search for data based on its attributes, its relationship with other data, and by its storage location.

Searching Data
Because of its ability to organize data based on various parameters, "WinFS" presents data as a directed acyclic graph (DAG) instead of as a tree (as in NTFS and FAT32). In NTFS and FAT32, since data is stored as a tree, it can be located based only on a single criteria (its directory path), while in "WinFS", since data is stored as a graph it can be searched based on multiple criteria.

Sharing Data
"WinFS" facilitates data sharing among various applications by giving file-backed data (documents, audio video, media, and so on) as well as non file-backed (e-mail, contacts, appointments, and so on), permanent residency status in the operating system and providing a unique API to access them. Now different applications can use these common pieces of data without worrying about maintaining individual data stores and the associated synchronization.

WinFS Building Blocks
Figure 1 illustrates the various building blocks of "WinFS".

Core "WinFS"
"WinFS" does not supplant NTFS; rather, it utilizes the basic file system services of NTFS. The core building block of "WinFS" includes the relational storage engine that provides file system services like ACL support, import/export, and quota management.

Data Model
"WinFS" defines a rich data model that resides on top of a relational storage engine. "WinFS" represents a piece of data (item) as a tuple in a relation. The attributes of the tuple describe the piece of data. Items can be related to each other by defining relationships between the tuples. "WinFS" also provides the ability to extend items and relationships.

Schemas
"WinFS" has built-in schemas to understand the rich metadata associated with your data. Some of the built-in schemas are for common data like documents, e-mail, contacts, appointments, tasks, and more. "WinFS" also allows you to write custom schemas for your own data.

"WinFS" provides certain services like synchronization and rules. These services are layered on top of core "WinFS" and provide "WinFS" with extended functionality. For example, the synchronization service enables you to synchronize two or more "WinFS" stores.

APIs
"WinFS"provides support for multiple programming models: object-oriented, relational, and XML based. "WinFS" can also be programmed using the Win32 API.

"WinFS" Type System Basics
"WinFS" is a "strongly typed" storage system. All data stored in "WinFS" is typed; that is, it is an object of some "WinFS" type. To understand the "WinFS" type system, we first have to understand the following four concepts:

• Items: All data is stored in WinFS  as a type specialized from Item (System.Storage.Item). Examples of types that derive from the Item are: Contact, Document, Task, and Event (all of which are found in the System.Storage.Core namespace)
• ScalarTypes: An atomic piece of information that describes an item.
• NestedTypes: A set of information that can be stored about an item. A nested type can have other nested types and scalar types in it.
• Relationships: Relate one item (source) to another (target). Relationships can be of three types:
  - Holding relationship: The source controls the lifetime of the target. These relationships are many-to-many, that is, a source can "hold" multiple targets and a target can be "held" by multiple sources. If a target looses all sources, it's deleted.
  - Reference relationship: Similar to a holding relationship but without the lifetime management of the target.
  - Embedding relationship: The source embeds the target. Strictly one-to-one.

Programming "WinFS"

In this section, we'll take a brief look at the object-oriented managed API provided by "WinFS". This API allows us to search, relate, and act upon data stored by WinFS

.An installation of Longhorn will have a single instance of "WinFS" service running on it. A "WinFS" instance can maintain multiple data stores. Each data source is referenced by its UNC path syntax as shown below:

\\<machine-name\<data-store-name>

All "WinFS" instances have a default store (DefaultStore). The default store on your "Longhorn" system will be called:

\\localhost\defaultstore

To program "WinFS", you have to get hold of an ItemContext object. This is done by calling the static Open method of the ItemContext class. The method is passed the UNC path of the store you want to program. If you pass nothing, the ItemContext refers to the default store.

All the types that are required to program "WinFS" can be found in three assemblies: System.Storage. dll, System.Storage.Schema.dll, and WinCorLib.dll.

As an example, Listing 1 is code that prints out all folders in the default store.

A very useful class provided by "WinFS" is the ItemSearcher (System.Storage.ItemSearcher) class, which allows you to search for items in a particular item type. Each item type (for example, Contact, Document, Folder, Event, and so on) has a static method called GetSearcher that takes in the ItemContext object and returns the ItemSearcher. Using the ItemSearcher, you can search for the items you need.

For example, Listing 2 uses the ItemSearcher to search for all documents whose title begins with the string 'Result'.

Conclusion
In this brief survey of "WinFS", we have seen that the future of data is with context-based storage systems. "WinFS" will provide the basis on which advanced data-mining tools are going to be built for the personal computer.

• Figure 1
• Source Code