Extensible Immersive Virtual Environments for Large Tiled Video Walls
Lorenz Cuno Klopfenstein
, Brendan D. Paolini
, Gioele Luchetti
and Alessandro Bogliolo
DiSBeF, Universit
a di Urbino “Carlo Bo”, Urbino, Italy
DII, Universit
a Politecnica delle Marche, Ancona, Italy
SAGE, IVE, Immersive Environment, Immersive Scenario, Virtual Environment.
The intent of the work is to present an Immersive Virtual Environment (IVE) as a new abstracted manage-
ment layer on top of the Scalable Adaptive Graphics Environment (SAGE) system, allowing the simplified
management and linear scaling of multiple SAGE-driven video walls, and the creation and control of simple
interactive scenarios. The framework exposes evolved Application Programming Interfaces (APIs) that are de-
tached from the underlying system and can be used by mobile clients as well. Primitives offered to developers
and content creators allow the definition of immersive cinematographic experiences using basic commands,
which are synchronized on the whole IVE environment. A complete implementation of the system is described
and then evaluated with the specific case of one physical installation.
High resolution immersive displays are becoming in-
creasingly common, for large scale interactive enter-
tainment, for “situation-room” displays, and for mas-
sive data visualization, where ultra-large images with
several millions of pixels are visualized on a single
coherent surface, including walls or monitor arrays.
Typical use-case scenarios include local and re-
mote groups of people working together on dis-
tributed and heterogeneous data, obtaining the ability
to control, manipulate, share, and visualize informa-
tion. Collaborative work on ultra-high-resolution dis-
plays with massive data has been found to be crucial
for several kinds of research and analysis (Renambot
et al., 2004).
As envisioned in different existing projects (Smarr
et al., 2003; Klosowski et al., 2002), large-scale visu-
alizations can neither be adapted to traditional desk-
top displays, nor to standard projection techniques.
Instead they usually require special setups with a
combination of very specific software and hardware.
For instance, when large scale visualizations are
not feasible with a single wide-Field Of View (FOV)
projector, with very expensive optics and hardware
(Zobel Jr et al., 1998), they are achieved using tiled
displays. That is, splitting the high-resolution im-
age into multiple separate regions, on different dis-
plays. This has been successfully adopted by using
non-overlapping projections (DeFanti et al., 2011),
side-by-side projections with precisely aligned over-
laps, or tiled LCD panels (Ni et al., 2006). Quite of-
ten, these installations are very expensive and difficult
to replicate.
A secondary issue of such large-scale visualiza-
tion systems, which can aggregate inputs from dif-
ferent sources and heterogeneous video data, is that
applications have to be re-designed or re-adapted in
order to be effectively used on that particular graphi-
cal environment, and they rarely take advantage of the
full graphical capabilities of a dedicated large visual-
ization system.
Most solutions in this field do not provide uni-
fied control of both input and output of the system.
The lack of higher abstraction interfaces for input
and output also entails a higher difficulty adapting
the setup to specific needs, which usually necessitates
very challenging integration work.
Some existing solutions, like LOTUS (Cho et al.,
2012), do indeed support predefined scenarios and
high-level input interfaces, but do not support mul-
tiple visualizations and dynamic reconfiguration.
The solution proposed by this paper is tailored for
multi-room entertainment systems or immersive inter-
active housing (i.e. multiple independent large-scale
visualizations that require central supervision). The
system is able to manage multiple concurrent visual-
izations, which are part of one overarching scenario,
and to reconfigure them dynamically.
Klopfenstein, L., Paolini, B., Luchetti, G. and Bogliolo, A.
Extensible Immersive Virtual Environments for Large Tiled Video Walls.
DOI: 10.5220/0005727803210328
In Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2016) - Volume 1: GRAPP, pages 323-330
ISBN: 978-989-758-175-5
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
The Scalable Adaptive Graphics Environment
(SAGE) is a cross-platform open-source middleware,
that was developed as a flexible graphics streaming
system. It enables users to compose multiple hetero-
geneous visualization applications seamlessly and
in real-time, on a very large display. Its decoupled
architecture allows rendering applications to take ad-
vantage of the processing power of the platform they
were developed for, without having to be constrained
or re-designed for SAGE’s graphic environment. At
the same time, compatibility with a large number of
data sources in a variety of resolutions and formats is
ensured (Renambot et al., 2004).
The primary aim of this software system is to
let local or distributed groups of users access, dis-
play and collaborate with ease on large datasets, in
a variety of resolutions and formats, which require
a big-scale visualization. Data may be generated by
high-performance storage and rendering clusters and
streamed over a high-bandwidth network in order to
be displayed on collaborative graphical end-points,
which may range widely in both surface area, reso-
lution and type of display (for instance a single high
resolution screen, a large tiled display or an array of
projectors). End-points provide means to efficiently
visualize data from multiple sources and to interact
with it.
SAGE has been put in use successfully in sev-
eral installations, including the CAVE (DeFanti et al.,
2011) and LambdaVision at the Electronic Visual-
ization Laboratory at the University of Illinois at
Chicago (EVL-UCI) (Renambot et al., 2005).
2.1 Architecture
SAGE is designed following a flexible and scalable
architecture, allowing multiple applications and video
sources to be streamed to a variety of displays, with-
out requiring any special hardware.
The system supports a single “scalable virtual
frame buffer” (a rectangular pixel matrix of almost
arbitrary size), which is tiled logically according to
the system’s configuration, matching the actual sys-
tem’s topology. One or more tiles, which make up a
contiguous screen area, are managed by a SAGE Re-
ceiver. It can receive a certain number of inbound
pixel streams, compose them and then push them to
the screens for display, as shown in Figure 1.
The Free Space Manager (FSM) acts as a window
manager for SAGE and collects and maintains the
information about receivers and other components.
The FSM orchestrates the configuration of the system
through a message passing protocol, directly com-
municating with other components and dinamically
reconfiguring the system when needed (Jeong et al.,
Pixel streams are generated by SAGE Applica-
tions. These are generic multi-purpose programs that
may run on any server of the system and make use
of the SAGE Application Interface Library (SAIL) to
provide image data to the SAGE Receivers.
Figure 1: SAGE architecture overview.
This distributed architecture allows the receivers
to be logically separated from the applications actu-
ally generating pixel data. Applications can run on
the same device of the target receiver, on a differ-
ent machine, or even on a remote machine connected
through a wide-area network.
Each single application thus contributes to the
tiled frame buffer, possibly pushing data through a
high-bandwidth network that bridges the computers
composing the SAGE system.
The FSM controls the pixel stream configuration
between applications and receivers: each application
is assigned to a specific region of the frame buffer
to draw to, which may change as applications are
started, terminated, moved across the frame buffer
and scaled in size. The FSM makes sure that each
application streams pixel data to the correct receivers
and that the pixels are thus drawn to the exact output
region on the frame buffer.
2.1.1 Applications and SAIL
The SAGE Application Interface Library (SAIL) pro-
vides a facility for programs to work as a SAGE Appli-
cation, by interacting with the FSM and by generating
pixel data to be shown on the frame buffer.
Data streamed through SAIL is encoded and trans-
mitted as raw pixels, usually in 24-bit RGB format or
using the lightweight lossless DXT compression for-
mat. Transmission occurs without any additional con-
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
version and without incurring in the cost and qual-
ity penalty of compression. While this also ensures
that the protocol has no strong platform dependency,
it makes it subject to very high bandwidth require-
SAIL represents a very thin layer between appli-
cation and network, that allows application develop-
ers to generate output pixels and transparently stream
them. The programming interface is built around a
very simple buffer-swapping mechanism: the applica-
tion generates raw pixel buffers and swaps them out
for SAIL to push them to the receivers. If necessary—
for instance when a SAGE Applications output region
spans across multiple output tiles—SAIL takes care
of splitting the pixel data and streaming it to the cor-
rect SAGE Receivers, according to the system’s con-
Existing programs, for most platforms and pro-
gramming languages, can easily be adapted to take
advantage of the large visualization capabilities of
SAGE thanks to this simplified streaming model. In
fact, SAGE by default provides a modified version of
MPlayer that is capable of working as a SAGE Appli-
cation by using a custom output video module.
2.2 Interoperating with SAGE
The SAGE FSM exposes a developer-facing inter-
face, called “Event Communication Layer”, that al-
lows SAGE UI modules and third-party controllers to
interact with the system.
Clients can send user messages to control the FSM
and the applications managed by it, for instance set-
ting up a new application, moving it from one location
of the frame buffer to another or resizing its output re-
gion. In return, clients obtain SAGE event messages
informing them about the current state of the system
and its running applications.
The layer is based on a low-level text message
exchange protocol on top of a TCP socket between
client and FSM. It supports high rates of message ex-
changes, but the messages use a basic format and are
limited to very simple information (see section 3.3).
2.3 Requirements
Large-scale collaborative visualization environments
intrinsically have very high requirements, for both
software and hardware.
In a SAGE-based system, when a SAGE Receiver
and an application using SAIL are not located on the
same machine, the amount of video data that must be
streamed through the network can be substantial. This
puts a very high performance requirement on the net-
work between nodes, which calls for high-grade mul-
tiple gigabit network equipment to be used.
Especially when reaching 4K resolutions or
higher, transmitting and handling video in real-time
puts a very high strain both on processing components
and the network. This can be exarcebated by the need
to handle different resolutions or encodings, and to
synchronize playback (Singh et al., 2004).
Finally, SAGE Receivers that individually drive
several displays require a custom multi-monitor setup
which usually also requires expensive video hard-
In the following section the Immersive Virtual Envi-
ronment (IVE) system will be presented. The system
is built on top of SAGE and presents a more extensible
way to coordinate immersive multisensorial scenarios
on one or more high resolution displays.
IVE is designed to control a multimedia interac-
tive scenario, taking place on multiple video outputs
and a variety of other sensorial actuators, like sur-
round audio systems, odor diffusers, etc.
As described in section 2.1, the SAGE system has
no higher abstraction understanding of the lifetime of
applications that draw on its frame buffer, nor of how
these applications interact with each other and other
components. IVE tackles this issue by wrapping one
or more SAGE instances and implementing a high
level API on top of them, thus enabling any client ap-
plication to describe, program and control complex
multimedia scenarios.
Figure 2: IVE architecture overview.
Extensible Immersive Virtual Environments for Large Tiled Video Walls
3.1 Architecture
The Immersive Virtual Environment (IVE) system is
designed to support multiple independent video walls.
IVE installations are split in logical groups, each one
of which is bound to a display surface that has no in-
tersections with surfaces of any other group.
IVE is structured in a centralized master/slave
topology as depicted in Figure 2: the whole system is
managed by a single supervising IVE Master, while
each independent display region (i.e., a logical group)
is controlled by an IVE Devil server. On its turn, each
Devil may control any given number of IVE Slave
servers, whose activities and interactions are still or-
chestrated by the master server.
Components of an IVE system are the following:
IVE Master: is a lightweight standalone server
implemented in Java, exposing a high-level com-
munication protocol. It manages the system state,
controls communication channels to the clients
and to groups of slave servers.
IVE Devil: overviews a single logical IVE group,
dedicated to an independent display surface. The
server receives commands and communicates
events back to the Master, thus keeping the sys-
tem’s state in sync. A Devil server internally
also runs a SAGE Free Space Manager instance
and controls it through its “Event Communication
Layer” (see 2.2), thus controlling the output to the
video wall.
IVE Slave: is a passive worker of a logical IVE
group. There may be multiple instances inside the
same group, each running a SAGE Receiver pro-
cess internally. A Slave also has the ability to run
any number of SAGE Applications. Local and re-
mote applications can stream video data to their
SAGE Receiver and thus contribute to the video
Generally, each component of IVE owns and controls
one or more SAGE components internally. Every IVE
logical group contains one Devil and any number of
Slave instances, which on their turn respectively man-
age the SAGE FSM (configured to work on the logical
group’s display), the Receivers and any Application
Which and how many instances of SAGE com-
ponents are handled by what IVE component is de-
scribed in Table 1.
On a large system, components will normally be
located on multiple machines in order to split the
load and to accomodate for physical distance between
components: a Devil node will control a certain num-
ber of Slaves installed on separate machines, depend-
Table 1: Relationship between IVE and SAGE components.
Free Space Manager (1)
Receiver (0-1)
Receiver (1)
Application (0+)
ing on the requirements of the installation. The total
number of machines needed mainly depends on the
size of the video wall and the covered surface’s ge-
The SAGE Receiver instance running on an IVE
Slave node can drive a number of displays (and cor-
responding tiles on the video wall) that is limited
only by its graphical hardware and, possibly, by soft-
ware limitations. Off-the-shelf consumer hardware
and software usually cannot run more than four dis-
plays at a time from a single workstation. These con-
straints aside, the whole output surface of a logical
group can be managed by a single SAGE Receiver in-
In fact, on small-scale systems all components can
in principle run on the same physical machine. Both
the role of the Devil and the Slave can be taken over
by the same node, which will also manage all SAGE
components required (the FSM, the Receiver and all
applications generating video data). Moreover, the
IVE Master itself can be run on the same machine as
well, in order to handle communication with client
3.2 IVE Management
All IVE components, Master, Devil and Slave, are
implemented in Java and run as standalone headless
processes running on a standard Java Virtual Machine
(JVM) instance.
Underneath the IVE software layer, IVE Devil and
Slave nodes will manage their respective SAGE pro-
cesses following the commands issued by the Master
node. All SAGE components run as standalone pro-
cesses, started and monitored by the IVE software:
this includes the FSM (managing the whole frame
buffer directly mapped onto the display surface of
that particular IVE group), SAGE Receivers (manag-
ing tiles of the frame buffer) and all applications gen-
erating the output video data.
When the IVE Master detects a new node coming
online (according to the procedure in section 3.4) it
opens a new bidirectional communication using TCP
sockets. Additionally, controller clients that manage
the IVE installation open a TCP communication chan-
nel to the Master node.
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
Communication between the IVE Master and
other parties occurs through a simple message ex-
changing protocol. Messages sent by the Master or by
a Controller client are called Commands, while those
sent in the opposite direction are called Events, up-
dating the Master or a Controller about a state change
occuring at a lower level.
IVE nodes support a variety of commands and
events, including:
Starting or stopping SAGE instances, shutting
down the system.
Starting, moving, resizing, rotating, terminating
or setting the transparency of applications. Video
players can start playback, pause or seek inside a
video file.
Notifications about a change of application state.
Messages are encoded as simple JSON objects. Each
message contains at least a “type” member with a
string value, describing the kind of command or event
it represents. Additional data is encoded as further
members of the message JSON object.
For instance, the command to resize an application
running on an IVE Slave is formatted as follows on the
appId: <SAGE APP ID>,
left: 0, right: 100,
top: 0, bottom: 100,
isFullscreen: false
In this case, the command sent by a Controller is
forwarded by the IVE Master to the IVE Devil man-
aging the application with a SAGE ID matching the
appId parameter. The IVE Devil then manipulates
the SAGE FSM to comply with the resize command,
as further described in section 3.3.
In addition to commands to start or manipulate ap-
plications, the IVE protocol also exposes a rich set of
APIs that allow Controller clients to query the state
of the system and present a coherent user interface
to the end-user. Any device capable of connecting to
the network and exchanging JSON messages through
TCP can be used to control the system.
By relying on the same Java code base of the IVE
servers, an Android application was developed, al-
lowing the user to easily and interactively control the
scenario using a tablet.
The message exchange protocol between IVE
nodes is very lightweight, requiring only the transmis-
sion of simple text messages, thus IVE nodes do not
need to be connected by a high-bandwidth network.
3.3 Interoperating with SAGE
A SAGE installation is composed by several indepen-
dent processes, that cannot easily be managed through
a single interface.
As described before, SAGE provides a simple
SAIL interface for applications to stream video data
to the output display (through a SAGE Receiver), but
this interface provides no coherent way to manage the
application itself or to interact with it. In fact, other
extensions to SAGE have been proposed in order to
overcome this limitation and to let users control ap-
plications (Fujiwara et al., 2011).
The default “Event Communication Layer” pro-
vided by the SAGE FSM exposes a simple text-based
interface to broadcast updates, in the form of strings
separated by newlines, and taking commands in a
similar form. The layer allows high communication
rates with low overhead. However the barebones for-
mat of the messaging protocol make the task of keep-
ing track of the system’s status burdensome.
SAGE messages have 4 fixed-size data fields and
1 payload field, all separated by null characters, fol-
lowing the layout seen in Table 2.
Table 2: Layout of a SAGE message.
Distance 8 bytes
Code 8 bytes
App code 8 bytes
Size 8 bytes
The Distance and App code parameters appear
to be generally unused. Message Codes are de-
fined by the SAGE documentation: commands are
in the 1000–1100 range, while event codes start from
40000. Payloads are expressed as a string of parame-
ters separated by ASCII spaces (Jeong et al., 2005).
For instance, the same resizing command gener-
ated by the IVE Master seen before will generate a
SAGE command on the IVE Devil running the appli-
cation as seen in Figure 3. The 1004 value repre-
sents the command code for resizing. Inside the pay-
load, the ID value will be replaced by the actual
SAGE application ID of the target application.
Application IDs are unique for each SAGE Ap-
plication in the context of a single FSM server. When
a new application process is spawned by a command
by the Master, the IVE Devil registers the applica-
tion to its FSM instance using a SAGE command with
code ‘1001’.
In order to keep the association between running
processes and application registrations up to date, all
IVE Devil instances also listen to application updates
Extensible Immersive Virtual Environments for Large Tiled Video Walls
Figure 3: Sample SAGE message of a resize command.
from the FSM process, which appear as SAGE mes-
sages with code 40001’. Any update to the applica-
tion’s state are surfaced through these event messages
and intercepted by the IVE process. These updates
contain information such as the unique ID assigned to
a newly registered application and its current position,
size and z-axis ordering.
At the same time, updates are also forwarded to
the IVE Master server, in the form of an IVE event
that contains a reference to the application, a refer-
ence to the IVE Devil, and the ID generated by the
FSM instance owning the application (i.e., control-
ling the frame buffer the application is contributing
to and where it is registered at). The Master server
keeps an updated map of all running applications and
their metadata, in order to uniquely identify running
processes and to track their activity.
How SAGE and IVE components exchange mes-
sages, commands and events is represented in Fig-
ure 4.
The server attempts to keep a coherent and com-
plete vision of the entire system, which goes be-
yond the limited vision that a SAGE server has about
its environment and the Application running therein.
For instance, the IVE Master can keep track of
whether a video player application is playing, paused
or stopped—based on its last control interaction—
and is able to provide a full snapshot of the system’s
Figure 4: Message routing.
state through its APIs. Commands, events and status
queries exposed by IVE enable a much more power-
ful and rich message exchange between the system
and its client controllers.
3.4 Auto Discovery
Being intended to work on dedicated installations,
SAGE mostly relies on static configuration files for
its setup. Instead, IVE is designed to work in an en-
vironment where subsets of the system may not be
available or may be intentionally turned off. More-
over, all components of an IVE installation are meant
to be located on the same network. Thus, IVE can
rely on an auto discovery process in order to detect
the available components as they come online.
System start-up works as follows:
1. IVE Master and Devil processes start as soon as
the boot phase of the systems where they respec-
tively reside is completed.
2. IVE Devil nodes announce their presence with pe-
riodical UDP broadcast messages on the network.
3. The IVE Master will detect newly started Devils,
register them and initiate a communication chan-
nel. Devils are setup by the Master according to
the configuration defined by the user during the
initial installation.
4. IVE Devils start their SAGE processes, including
the FSM and the Receiver. These SAGE configu-
rations are highly tailored to their hardware setup
and depend on preset local configuration files.
5. The Master node will then load the preconfigured
scenario, starting applications by sending com-
mands to the Slaves, and begin receiving events
from them in order to maintain track of the sys-
tem’s state.
3.5 Load Balancing
A SAGE Application can be drawn on any region of
the video wall, sometimes overlapping two or more
tiles which are controlled by different SAGE Re-
ceivers. When the application and the receiver are lo-
cated on different physical machines, some part of the
video data generated by the application must be trans-
ferred through the network in order to reach the video
wall. Since SAGE mostly transfers raw video data,
this may put a very high strain on the network, quickly
reaching gigabit-per-second bandwidths (Singh et al.,
IVE is designed to keep a high level overview of
where applications run and where they are displayed.
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications
Thus, the IVE Master node takes into account the tar-
get region where an application will be shown and,
when starting a new SAGE Application process to
generate video data, will spawn the process on the
nearest machine to the Receiver (ideally on the very
same machine).
As seen in Figure 5, when an application is run-
ning on the same IVE Slave as its receiver, all data
transfer through the network can be avoided. When a
part of the target region, albeit small, overlaps on an-
other tile of the video wall, the video data must be
streamed from the application to the corresponding
receiver, which requires a high-bandwidth data trans-
Figure 5: Application load balancing.
At the moment, the IVE Master only takes into
account the initial starting region where an applica-
tion will be drawn when selecting a machine to spawn
the process. If an application is moved or scaled, its
draw region may move to another tile. In this case,
relocating an application process from one machine
to another is possible for simple, static applications
(e.g., image viewers), but still poses a challenging
problem for applications like videos or rendered out-
puts, which would need to transfer their state and re-
sume seamlessly on another machine.
A prototypal installation of the IVE system has been
realized in a dedicated room for demonstration and
testing purposes, shown in Figure 6. The main wall of
the room measures about 11 × 3 m, while secondary
side walls measure 2 × 3 m.
The setup is intended to entirely cover the main
surface and the lateral walls of the room with a sin-
Figure 6: IVE installation running an immersive scenario.
gle complex visualization. To this purpose, 8 BenQ
LW61ST projectors where used for the main surface
and 2 identical projectors for each side wall, totaling
12 projectors to cover the entire surface. The overall
projected frame buffer accounts for 7680 × 1440 total
pixels (roughly 11 mega pixels).
The surface is split in three logical regions, each
managed by a single IVE Slave node. The workstation
managing the main screen also runs the IVE Master
and a Slave server at the same time. (The adopted
setup makes IVE run as a standalone system, as a sin-
gle logical group, however the IVE Master may in fact
be located outside the room or even remotely, while
also managing other IVE Devils at the same time.)
All IVE nodes are connected by a high-
performance 10 gigabit network
, which supports
high-bitrate raw frame buffer data passing from one
server to another. The network makes use of one Net-
gear Prosafe XS708E gigabit ethernet switch and
workstations are equipped with Intel X540-T1 net-
work cards.
The central workstation running the Master and
the main Devil instance has the more onerous work-
load, having to drive 8 projectors while also con-
trolling the other two machines. The workstation
is equipped with two NVIDIA Quadro K5000 video
cards (each sporting 2 DVI and 2 Displayport video
outputs) and an NVIDIA Quadro Sync card. The
two secondary computers driving the 4 projectors on
the side rooms are equipped with single NVIDIA
GTX670 video cards (providing 2 video outputs
each). Both kinds of workstation are also equipped
with Intel i7 3770 processors, SSDs and respectively
8 and 4 GB of RAM.
This kind of high-bandwidth network was found to be
necessary for the playback of 4K videos. For lower resolu-
tion video sources, a 1 gigabit network would suffice.
Extensible Immersive Virtual Environments for Large Tiled Video Walls
Each machine runs Ubuntu 12.04 LTS, using the
Compiz desktop manager and the LightTwist plug-in
for projection deformation correction and alignment.
The proposed IVE system provides an extensible
management solution for complex video installations,
composed of one or more large-scale video walls
driven by SAGE.
The IVE system is capable of managing multiple
SAGE instances, hiding the complexity inherent in
managing and interacting with SAGE processes and
applications through a coherent front-end that keeps
track of the underlying system’s state. All of IVE’s
features are accessible through high-level APIs that
can be used by any software client, including mobile
IVE also provides several extension points, pos-
sibly including interfaces for other systems and pro-
tocols (including KNX for domotic devices, lighting,
odor diffusers, etc). Interactive or passive scenarios
can be implemented through the programmable APIs.
5.1 Future Work
In the near future, the adoption of multimedia instal-
lations made possible by IVE will expand beyond the
usage in academic institutions and art installations,
eventually reaching offices and private homes. Such
large-scale visualizations will not only be limited to
collaborative work, but also include immersive enter-
tainment, multisensorial experiences and even more
passive uses, like digital wallpapers.
Such scenarios shall be explored, providing new
use-cases for the adoption of IVE other than large data
visualization and entertainment. In fact, the prototype
used to test IVE has been adapted and will be tested
in a interactive multisensorial housing installation.
Furthermore, since IVE relies on a large number
of distributed components, the possibility of running
the whole system or parts of it on low-power embed-
ded systems will be explored. In particular, machines
running specific applications or the SAGE FSM could
be replaced by a cluster of embedded devices that can
be efficiently managed by the supervising IVE Master
Cho, Y., Kim, M., and Park, K. S. (2012). Lotus: compos-
ing a multi-user interactive tiled display virtual envi-
ronment. The Visual Computer, 28(1):99–109.
DeFanti, T. A., Acevedo, D., Ainsworth, R. A., Brown,
M. D., Cutchin, S., Dawe, G., Doerr, K.-U., Johnson,
A., Knox, C., Kooima, R., et al. (2011). The future of
the CAVE. Central European Journal of Engineering,
Fujiwara, Y., Ichikawa, K., Takemura, H., et al. (2011).
A multi-application controller for SAGE-enabled tiled
display wall in wide-area distributed computing envi-
ronments. Journal of Information Processing Systems,
Jeong, B., Jagodic, R., Spale, A., Renambot, L., Aguilera,
J., and Goldman, G. (2005). SAGE documentation.
SAGEdoc.htm. Accessed: 2015-09-28.
Jeong, B., Renambot, L., Jagodic, R., Singh, R., Aguil-
era, J., Johnson, A., and Leigh, J. (2006). High-
performance dynamic graphics streaming for scalable
adaptive graphics environment. In SC 2006 Confer-
ence, Proceedings of the ACM/IEEE, pages 24–24.
Klosowski, J. T., Kirchner, P. D., Valuyeva, J., Abram, G.,
Morris, C. J., Wolfe, R. H., and Jackman, T. (2002).
Deep view: high-resolution reality. Computer Graph-
ics and Applications, IEEE, 22(3):12–15.
Ni, T., Schmidt, G. S., Staadt, O. G., Livingston, M., Ball,
R., May, R., et al. (2006). A survey of large high-
resolution display technologies, techniques, and ap-
plications. In Virtual Reality Conference, 2006, pages
223–236. IEEE.
Renambot, L., Johnson, A., and Leigh, J. (2005). Lamb-
daVision: Building a 100 megapixel display. In
NSF CISE/CNS Infrastructure Experience Workshop,
Champaign, IL.
Renambot, L., Rao, A., Singh, R., Jeong, B., Krishnaprasad,
N., Vishwanath, V., Chandrasekhar, V., Schwarz, N.,
Spale, A., Zhang, C., et al. (2004). SAGE: the scal-
able adaptive graphics environment. In Proceedings
of WACE, volume 9, pages 2004–09. Citeseer.
Singh, R., Jeong, B., Renambot, L., Johnson, A., and
Leigh, J. (2004). Teravision: a distributed, scalable,
high resolution graphics streaming system. In Cluster
Computing, 2004 IEEE International Conference on,
pages 391–400. IEEE.
Smarr, L. L., Chien, A. A., DeFanti, T., Leigh, J., and Pa-
padopoulos, P. M. (2003). The OptiPuter. Communi-
cations of the ACM, 46(11):58–67.
Zobel Jr, R. W., Bennett, D. T., Idaszak, R. L., and Ko-
vach, D. (1998). Multi-pieced, portable projection
dome and method of assembling the same. US Patent
GRAPP 2016 - International Conference on Computer Graphics Theory and Applications