Scalable Fabric: Flexible Task Management
George Robertson, Eric Horvitz, Mary Czerwinski,
Patrick Baudisch, Dugald Hutchings, Brian Meyers, Daniel Robbins, and Greg Smith
Microsoft Research
One Microsoft Way
Redmond, WA 98052 USA
{ggr; horvitz; marycz; baudisch; brianme; dcr; gregsmi}@microsoft.com; hutch@cc.gatech.edu
ABSTRACT
Our studies have shown that as displays become larger, users
leave more windows open for easy multitasking. A larger number
of windows, however, may increase the time that users spend
arranging and switching between tasks. We present Scalable
Fabric, a task management system designed to address problems
with the proliferation of open windows on the PC desktop.
Scalable Fabric couples window management with a flexible
visual representation to provide a focus-plus-context solution to
desktop complexity. Users interact with windows in a central
focus region of the display in a normal manner, but when a user
moves a window into the periphery, it shrinks in size, getting
smaller as it nears the edge of the display. The window
“minimize” action is redefined to return the window to its
preferred location in the periphery, allowing windows to remain
visible when not in use. Windows in the periphery may be
grouped together into named tasks, and task switching is
accomplished with a single mouse click. The spatial arrangement
of tasks leverages human spatial memory to make task switching
easier. We review the evolution of Scalable Fabric over three
design iterations, including discussion of results from two user
studies that were performed to compare the experience with
Scalable Fabric to that of the Microsoft Windows XP TaskBar.
Categories and Subject Descriptors
H.5.2 [Information Interfaces and Presentation]: User
Interfaces – graphical user interfaces, windowing systems.
General Terms
Management, Design, Experimentation, Human Factors.
Keywords
Task Management, Scaling, Interaction, Spatial Memory.
1. INTRODUCTION
Twenty years ago, Bannon et al. [1] observed that information
workers often switch between concurrent tasks or activities. In
Rooms, Card and Henderson [6] observed that tasks can be
supported via the management of “working sets” of windows, in
much the same way operating systems manage working sets in
memory. Card and Henderson identified desirable properties of
task management systems, including: fast task switching, fast task
resumption, and easy reacquisition of the cognitive context
associated with a task.
Over the two decades since the work of Bannon et al., numerous
virtual desktop managers have been built and each has exhibited
some of these properties. Task management systems typically
provide some efficient way of switching from one set of windows
and applications to another set, as a basic form of task switching.
Although workers may switch among tasks in a self-guided
manner, a significant portion of task switching is caused by
external interruptions [12]. Czerwinski, Cutrell, and Horvitz
[7][9][[10] have sought to understand the influence of
interruptions on task switching for information workers in order
to design tools that can assist users to recover from interruptions.
We have also been motivated to re-examine task switching and
task management design opportunities in the face of the growing
popularity of larger display and multiple monitor configurations.
In an informal study at our organization, we found that when
users shift to larger display surfaces, they leave more applications
running and associated windows open. For example, we observed
that single display users tend to keep an average or 4 windows
open at once, while dual monitor users keep 12 and triple monitor
users keep 18 windows open on average (N=16 users). This
significant trend suggests that there is an opportunity for design
innovation with windows and task management to make handling
larger numbers of concurrent windows, potentially clustered by
task, a fundamentally more natural and effective experience.
We have developed a windows management methodology to
exploit this opportunity. Scalable Fabric is a system designed to
assist users manage tasks on the Windows desktop, allocating
screen real estate in accordance with a user’s attention, using a
focus-plus-context display. The periphery of the screen is used to
hold scaled down live windows rather than hiding them with
traditional windows minimization. In order to facilitate task
switching, Scalable Fabric allows users to group collections of
windows that are used together. We shall refer to groups of
Windows that are used together as tasks. We realize that this
notion is not isomorphic with all conceptions of computer-centric
“tasks,” but in our conversations with end users after studies of
this topic, this notion appears to resonate easily with their
description of their own computer work.


85
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page. To copy
otherwise, to republish, to post on servers or to redistribute to lists,
requires prior specific permission and/or a fee.
AVI '04, May 25-28, 2004, Gallipoli (LE), Italy
© 2004 ACM 1-58113-867-9/04/0500..$5.00

In the remainder of the paper, we will first discuss related
research. Then we will describe details of the Scalable Fabric
methodology. We present the results of a comparative user study
of Scalable Fabric and the Windows TaskBar, and a longitudinal
field study of Scalable Fabric. Finally, we discuss project
directions and opportunities for future research.
2. RELATED WORK ON TASK
MANAGEMENT
The most popular software system for task management is the
virtual desktop manager. One of the earliest designs exploring a
virtual desktop manager was Smalltalk Project Views [13]. Rooms
[6][15] is probably the most well-known of these kinds of
systems. A number of virtual desktop managers are currently
available, and are described in [27]. We have not been able to find
evidence that these systems have been evaluated in a formal
manner. Thus, it is difficult to ascertain how easy they are to use
or how well they integrate into real-world settings.
In addition to virtual desktop managers, a number of novel
solutions have been proposed, including extending the user’s
desktop with additional low-resolution screen space [2],
employing 3D environments as pursued by the TaskGallery [24]
effort, providing a zoomable space as in Pad++ [3], and the use of
time as the main axis and organizing principle [22]. Also, tiled
window managers [5][26] have been created to address some of
these same issues, as well as systems that involve the invocation
of bumping processes among windows, to allow a window at
focus to push others away [4][16].
We have pursued prototypes of temporal and spatial visualizations
of users’ daily computing configurations. These designs use
lightweight, temporal cues, such as the state of a user’s desktop at
different times [19]. We have also sought to provide support for
task-based visualizations and switching, in a similar vein to the
work of Henderson & Card [15], Kaptelinin [17], Macintyre et al.
[18] and Robertson et al. [24].
In distinction to the prior work, we have explored designs for
virtual desktop organizers that do not replace the entire PC
desktop with a new metaphor, but rather occupy the same
conceptual and physical space that is already devoted to window
management in the Windows OS – namely, the area in the
periphery of the display surface. Using these prototypes, we have
been performing longitudinal studies on the benefits of temporal
and visual cues for enhancing memory about knowledge-based
tasks, in order to facilitate task switching. We seek to understand
the potential benefits from the use of these systems, and to iterate
their design. For example, the Windows XP TaskBar provides
“grouping by application” to address the problem of running out
of bar space, e.g., all Word windows are grouped together, and all
Internet Explorer windows are grouped together. Grouping by
application, rather than by task, can create user confusion, as
specific windows executing the same application may be
conceptually unrelated to each other, and cross-application
windows may be used together on one user activity [12].
We also address the challenge of accessibility of windows
belonging to different tasks. While virtual desktop managers
typically impose strict separation between tasks, we allow users to
simultaneously display any subset of windows, even if they
should be assigned to different tasks. Rooms’ placements
mechanism allows a window to appear in multiple virtual
desktops, but this requires forethought to set up. The approach in
Scalable Fabric is more dynamic and requires no forethought.
In related work, GroupBar [25] addresses these latter issues by
evolving the Windows TaskBar to support task groups of
windows on a bar, using the same minimized window
representation used by TaskBar.
Although GroupBar has most of the properties we were seeking in
a task management system, the design does not effectively
leverage human spatial and visual recognition memory. We know
from user studies on the Data Mountain [23] and Task Gallery
[24] that spatial memory works in a virtual environment similarly
to the way it works in the physical world, and that user task
performance is enhanced, particularly when the task involves
retrieving items placed spatially. GroupBar makes limited use of
spatial memory by allowing users to create multiple bars.
Limitations stem from the bar design, which is linear, list-based,
and does not expose much virtual space in which to place tasks.
Scalable Fabric makes use of the periphery of the display for
spatial layout of tasks, in addition to leveraging users’ efficient
visual recognition memory for images [8]. Scalable Fabric allows
users to leave windows and clusters of windows open and visible
at all times via a process of scaling down and moving the
windows and clusters to the periphery. This idea was partially
inspired by observations we made with Data Mountain; items
toward the back of the Data Mountain take much less space, but
are still readily recognizable. It was also inspired by the scaling at
the edges of the display in Flatland [21] and by ZoomScapes’
location based scaling mechanism [14]. While ZoomScapes is not
a task management system, its management of sheets and groups
of sheets is similar to Scalable Fabric’s management of windows
and tasks. We shall review the differences in the next section.
3. SCALABLE FABRIC BASICS
In Scalable Fabric, the user defines a central focus area on the
display surface by moving periphery boundary markers to desired
locations. In Color Plate 1, these boundary markers are visible,
but users usually hide the boundary markers unless they are
changing the size or shape of the focus area, in which case the
markers serve as resize handles. Within the focus area, windows
behave as they normally do in the Windows desktop. The
periphery contains windows and collections of windows (or tasks)
that are not currently in use, but may be put to use at any moment.
Windows in the periphery are smaller so that more tasks can be
held there when the user is focusing on something else. With this
metaphor, we believe users will rarely need to close or minimize
windows in the traditional sense. Users can take advantage of
extra screen real estate, especially on larger displays, to allow the
peripheral windows to always be visible.
When a user moves a window into the periphery, it shrinks
monotonically with distance from the focus-periphery boundary,
getting smaller as it nears the edge of the screen. When the user
clicks on a window in the periphery, it returns to its last focus
position; this is the new “restore” behavior, and is accomplished
with a one second animation of the window moving from one
location to the other. When the user “minimizes” a window in the
focus area, e.g., by clicking the window’s ‘minimize’ button, it
returns to its last peripheral position.
86