Access Grid Nodes

The Access Grid is a project initiated by Argonne National Laboratories, Maths and Computer Science, Futures Laboratory in the USA. It is essentially an open global project to develop a large scale collaborative environment, similar to video-conferencing rooms today but scaled up dramatically in flexibility and functionality.

The IF group is involved in building 4 Access Grid nodes at the ANU today:

- AG@CSIT: [50 seats] N101 Seminar Room, CS&IT Building
- AG@Huxley: [8 seats] Room 318, Leonard Huxley Building
- AG@Stromlo: [95 seats] Duffield Lecture Theatre, Mt Stromlo Observatory
- AG@Baume: [50 seats] Lecture Theatre, Peter Baume Building, ANU Medical School

Other nodes are being planned at ANU. More details on these three nodes will be available through links above shortly.

There are currently around 300 AG nodes around the world, with about 20 in Australia. This is growing rapidly, with an expected doubling of nodes in Australia and internationally by mid-2005.

Background - What is an Access Grid node?

In "normal" room-based videoconferencing, the most common standard network protocol is called H.320 for ISDN-based calls and H.323 for Internet-based calls. These standards come out the telecommunications industry, and work extremely well. There are now millions of H.323 devices around the world, ranging from large room systems down to small units for personal use, that you can attach to your laptop computer.

The disadvantages of H.320/H.323 are that they come from an era of limited bandwidth, and so they are very conservative in what they transmit. This makes them very network friendly, but often not very application friendly.

- Typically you are limited to one active camera at a time, which is a problem when you want to show a presenter and their slides and their audience at the same time.
- They are designed for use between two sites. If you want to have 3 sites participating you require additional hardware (an "MCU"), which can be very expensive
- The quality of the audio and video is determined by the bandwidth, and the algorithms used to encode the transmission. At this stage the integration of new algorithms in H.323 products has been very slow, and the algorithms are targeted at low bandwidth and hence lower quality, compared to say broadcast quality TV or DVD video.
- The ability to share applications, such as slideshows, web browsers, animations, simulations, etc. is very restricted. Again, the bandwidth is one issue, but also there are very limited ways in which applications can be generically "shared" across a network.
- There is almost no capacity to integrate more novel devices, such as electronic whiteboards, pen trackers, 3D visualisation environments, robotic devices and so on.

Having said that, H.323 is extremely stable and well developed, has a good market base, and products are generally very user-friendly. Their telecommunications heritage means they integrate well with other similar protocols, such as telephony, ISDN and satellite networks.

So what does the Access Grid provide?

The Access Grid tries to turn the above on its head. It places no constraints on bandwidth, nor on encoding algorithms. This allows multiple cameras to be used simultaneously, and at higher quality. It uses network-based services (IP multicast) to deliver traffic to multiple sites, without additional hardware required. It is trying to develop ways of sharing applications that are more tuned to the requirements of each application. There are also several projects underway to link in more novel devices.

All of the software is freely available, and all of the hardware is commercial off-the-shelf, and readily available. This means it is easy to adjust each AG node, in hardware terms, to suit each situation, and it is relatively easy to add your own software to it.

A proper Access Grid node is normally a room with a very large-scale display, with associated computing and audio/video hardware, from 8-100 seats at the first three ANU nodes. However, it can be run on a desktop machine, a laptop, and even on a handheld device such as a PDA. One gives up on performance and flexibility in such a setup, but it does allow participation in larger events.

Having set out to do all of this though, it needs to be noted that the Access Grid nodes are not (as) widespread (yet), their user-interface is less than friendly, the protocol standards are still very basic, and their overall robustness is suspect.

This is where the IF group comes in. We are developing not just the skills to build and operate AG nodes for users, but are seeking to dramatically enhance their functionality, usability, and reliability. This not just for our use alone but for all AG nodes.

IF Access Grid related projects

Benchmarking: RTPreplicate

We are developing tools that allow developers to benchmark the performance of their hardware used in their AG node, which in turn allows us to specify more precisely the best hardware for a node in any situation. The first tool is rtpreplicate. This tool receives an incoming multimedia stream and replicates it any number of times before forwarding it on to the machine(s) under test. Most systems don't cope well with 300 simultaneous incoming video streams... yet!

Connectivity: rcbridge (RC: remote-control), and rcb-forward

The Access Grid uses IP multicast for its underlying network transport. Unfortunately, multicast is not always easy to deploy, nor to debug. Various solutions exist for "tunneling" the traffic over normal links, but they are not scalable nor user-friendly at either end of the tunnel. We have developed rcbridge (RC: remote-control), which is a multicast tunneler, with a web interface for remote control. This allows a multicast-enabled site to share its connectivity with any number of sites it chooses, and doesn't have to manually configure it each time. The receiving end can also select which multicast streams it wants to receive, allowing sites with limited bandwidth to participate, e.g. those at home or on wireless links. This system works in both directions, so that your outbound audio/video is also visible to everyone else.

There is a matching rcb-forward tool that can be run on your local network, that talks to an rcbridge service and basically transparently "tunnels" multicast across the network between them, without requiring other applications on the subnet to be modified or reconfigured. This tool will be integrated into the AG environment.

User interface: VP

We have started work on a replacement tool for the AG's existing audio and video clients, which are now nearly 10 years old. The tool is called VP. It will provide several new, and extremely useful features for the Access Grid (and as a stand-alone application). The major enhancements revolve around

- the user-interface, reducing the need for a person to run each node on site
- the audio subsystem, providing lip-synch between audio and video, as well as audio-localisation across the AG nodes' large display system
- performance enhancements, with faster and higher-quality video algorithms, multiple inputs/outputs, and support for distributed encoding/decoding
- reliability enhancements, making the systems more robust
- a new capture application, taking advantage of new hardware, drivers, codecs as well as providing more complex metadata to the VP display application to enhance the user interface. This subproject is supported by a grant from the Apple University Developer Fund (AUDF), will be developed on the Mac platform and will also see VP ported to MacOSX.

Integration: Silver323

Ultimately, the world is still largely using H.323 products for videoconferencing. There are currently only very rudimentary ways of bridging the AG world into the H.323 world. The IF group has hired a team of software engineering students to tackle this problem. They are developing Silver323 as a gateway service, which will run on a multicast-aware server, and allow remote H.323 users to connect and participate. There are many interesting challenges, in the network standards and user interface differences between the two worlds. We hope Silver323 will provide a very flexible solution to many of these challenges.

This will bring the benefits of H.323 to the Access Grid, e.g. allowing ISDN-connected sites to participate, as well as ordinary telephone devices on any phone network in the world. For now, the phones will only receive and transmit audio, but already we are seeing phones with rudimentary video capabilities on the market. Other extensions being considered include support for SIP and RTSP

Recording: Remora(?)

An AccessGrid "stream" is more complex than normal audio/video archives, in that they contain many streams, from multiple sources, and they can also carry (potentially) additional application data streams, such as slide show events and whoteboard events. The existing rtpdump approach does not scale well, and does not record all of the potential material.

A project in 2004 will develop a new recording format, based around packet events similar to rtpdump, but with more efficient indexing and metadata tracks, support for non-rtp-based content streams, and links to external metadata such as annotations.

A recording tool/service will also be created, which can be linked to the AG display application via mbus, allowing for recordings to be created quickly on-demand from within the user interface. A post-production tool will then be built to edit an AGar recording into leaner files, or different formats (e.g. single-video mpeg2 streams). The whole toolkit will remain small enough to install on a local machine, but could be scaled to much larger architectures.

Hardware:

A typical AG node contains hardware as diverse as cameras, projectors, recorders, and electronic whiteboards. A project in 2004 will develop an AG client interface for these kinds of hardware (RS232, USB, etc.) to allow remote control for node operators or remote participants, to stream whiteboard events, and to provide some additional security of node hardware.