search:

10   Virtual Collaborative Environments

Virtual Collaborative Environments is an umbrella activity for a number of projects that use, in one way or another, multimedia transmission over high-speed networks. The whole activity can be divided into two major areas:

• synchronous communication infrastructure designed for interactive communication such as videoconferencing
• asynchronous tools with relaxed latency requirements, which often use unidirectional data transmission only.

In both areas, our work ranges from a pure theoretical research through practical implementations to a support infrastructure for production collaborative environments.

10.1   Synchronous Communication Infrastructure

The problem of synchronous (on-line or interactive) processing lies in providing an environment with as low a latency of processing and distribution as possible. We are focusing on the development of user-empowered network support for synchronous multi-point distribution that is both scalable to a large number of clients and also robust with respect to outages of network links and other network elements. This research is a follow-up to our development of active network elements (routers, reflectors etc.).

10.1.1   Active Elements

Real-time virtual collaboration needs a synchronous multimedia distribution network with high capacity and low latency. Such a network can be composed of interconnected service elements - so called active elements (AEs) [HHM05]. They are a generalisation of the user-empowered programmable reflector described in [HHD04]. The reflector is a programmable network device that replicates and optionally processes incoming data, usually in the form of UDP datagrams, using unicast communication only. If the data is sent to all the listening clients, the number of data copies is equal to the number of the clients, and the upper bound for egress traffic volume grows with n(n-1), where n is the number of sending clients. It runs entirely in user space and thus works without any need for administrative privileges on the host computer, which a prerequisite for the user-empowered character of the device. The design of the reflectors and AEs is based on our active router blueprint [HlS01], uses the same principles of modularity and, in addition, applies the user-empowered approach.

Active elements add capabilities for inter-element communication and also for distributing their modules over a tightly coupled cluster. Only the communication capability is important for scalable environments discussed in this report, while the research on distributed AEs was only started in 2005. Network management is implemented via two modules dynamically linked to the AE at run-time: Network Management (NM) and Network Information Service (NIS) as shown in Figure 10.1. The NM module takes care of building and managing the network of AEs, joining new content groups and leaving old ones, and reorganising the network in case of a link failure. NIS collects and publishes information about the specific AE, such as its available network and processing capacity, about the network of AEs, about properties important for synchronous multimedia distribution (e.g., pairwise one-way delay or round trip time, estimated link capacity), and also information on the contents and available formats distributed by the network.

For out-of-band control messages, the AE network uses self-organising principles already successfully implemented in common peer-to-peer (P2P) network frameworks, namely for discovering AE neighbours, available services or content, topology maintenance, and also for control channel management. The P2P approach satisfies the requirements on both robustness and user-empowered character. Its lower efficiency has no significant impact as it routes administrative data only. A prototype was implemented [HoH05] using the JXTA P2P substrate [JXTA].

10.1.2   Re-balancing and Fail-Over Operations

The topology and usage patterns of any network change rather frequently, and these changes must be reflected in the overlay network, too. We consider two basic scenarios:

1. re-balancing is scheduled as a result of either a change in usage patterns or introduction of new links and/or nodes, but not as a result of a link or AE failure
2. a reaction to a sudden failure

In the first scenario, the infrastructure re-balances to a new topology and then switches to it for sending data. On the contrary, a sudden failure in the second scenario is likely to result in packet loss (for unreliable transmission protocols like UDP) or delay (for reliable protocols like TCP), unless the network distribution model has some permanent redundancy built in. The probability of a failure of a particular link or AE is rather small, despite high frequency of network failures that can be observed on a large scale. Therefore, double redundancy might be sufficient for the majority of applications, perhaps with an option to increase it for the most demanding applications.

10.1.3   Network of Active Elements

Separation of the control plane from the data distribution plane allows for a modular implementation of distribution models with different properties. We studied a number of data distribution models for AE networks [HHM05] with different performance and robustness properties:

2D full mesh

the simplest model with very high robustness; an AE outage only influences the clients that are directly connected; it also minimises number of hops inside the network

3D layered mesh

this model provides an improved performance over the 2D model while retaining the recovery behaviour and minimum number of hops inside the network

3D layered mesh with intermediate AEs

additional improvement over the 3D layered model, which can be seen as a transition to spanning trees

redundant (minimum) spanning trees

a model allowing maximum flexibility, efficient recovery from network outages and optimised data distribution with respect to the saturation of lines; extension to multiple redundant spanning trees brings an additional benefit of a very fast recovery

10.2   High-Definition Interactive Collaborative Environments

Enabled by current high-speed networks, high-definition (HD) video transmissions have become an essential tool for many applications. A truly interactive collaborative environment is still a serious challenge, as data processing must be severely limited (ideally to less than 100 ms) in order to achieve an acceptable level of interactivity. Consequently, uncompressed video is the most convenient option, which however imposes high demands on the underlying networking infrastructure, especially for multi-point data distribution, as each video stream has 1.5 Gbps. During 2005, we developed and successfully demonstrated a prototype of a low-latency multi-part collaborative environment based on uncompressed HD-SDI video complying to SMPTE 292M, which was also used as a transport medium for distributed collaborative visualisation. During the demonstrations, the whole system turned into a very interesting example of integration of high-end applications of optical networking infrastructure, as shown below. Detailed information on both the system and the demos can be found at https://sitola.fi.muni.cz/igrid/.

The whole system consists of two basic parts: client applications and network distribution and processing. We developed the client tools based on the DVS Centaurus and Chelsio 10GE cards and the UltraGrid software by Colin Perkins [PGL02]. The latter software was extended to support full 1080i HD (previously only lower 720p resolution was supported) and full software display. As the Centaurus card can also be used for video display, we only extended UltraGrid by adding support for software-only display, including field de-interlace algorithm and colour space downsampling from 10 to 8 bits per colour plane to relax the requirement of expensive DVS Centaurus cards being on both ends of a HD video path. The computation-intensive parts of the UltraGrid software were optimised at the assembly language level for use on the AMD64/Opteron-based computers. The total end-to-end latency in a laboratory setup with both computers connected to the same 10GE Cisco Catalyst 6506 switch, was 175+/-5 ms and we are actively working on reducing it even further.

The network distribution requires a multi-point distribution service in order to deliver data from each participant to all others. As our previous experience with IP multicast over heterogeneous networks was not satisfactory, we decided to rely on the Active Element technology described above. We demonstrated its usability on 10GE networks where each AE was able to duplicate a 1.5 Gbps stream on a commodity dual-AMD64 PC from one partner to two other partners.

10.2.1   iGrid 2005 Demonstration

The iGrid workshop is a bi-annual event where teams from all over the world demonstrate most advanced applications of lambda services and high-speed IP networks. At iGrid 2005, which took place in in San Diego in September 2005, CESNET participated in two demonstrations: CZ101 (HD Multi-point Conference) organised by CESNET and US127 (Interactive Remote Visualisation across the Louisiana Optical Network (LONI) and the National LambdaRail) organised by Louisiana State University (LSU). In CZ101 we demonstrated a low latency collaborative environment involving partners from CESNET and Masaryk University in Brno, LSU in Baton Rouge and iGrid premises in San Diego. Sending 1.5 Gbps from each site means that each site has to receive 3 Gbps traffic from the other two sites and thus the aggregated traffic volume in both directions is 4.5 Gbps. Multi-point distribution was handled by three AEs located in StarLight (Chicago), where the lambda services from all the three sites were terminated. In US127, remote visualisation data from Baton Rouge were transmitted and distributed as uncompressed HD video over IP using the same AEs in Chicago.

These demonstrations required not only an intensive collaboration within CESNET, in particular between the activities Virtual Collaborative Environments and Optical Networking, but also a very tight coordination with our networking partners from SURFnet, StarLight and National LambdaRail. The demonstrations proved that current high-end networks based on lambda services are very suitable for this kind of network applications, since we were able to transmit 4.5 Gbps traffic sensitive to jitter and loss without any serious problems.

Our demonstrations at iGrid were accepted very positively by the international community and we were invited to participate in two subsequent events. The first one, which took place few days after iGrid, was a demonstration of distributed interactive visualisation to Dr. Jack Marburger, a scientific advisor to U.S. president G. W. Bush.

10.2.2   SuperComputing 2005 Demonstration

The second demonstration we were invited to participate in was another distributed interactive visualisation organised by LSU during SuperComputing 2005, which took place in November 2005 in Seattle, USA. Like US127, the visualisation data were generated at LSU in Baton Rouge and distributed over the national LONI network to all three participating sites. All the sites were able to control the visualisation by adjusting certain parameters interactively using knobs (see Figure 10.3). As the demo also included transmission of video and audio from the remote sites, the participants were able to discuss and analyse the visualisation in real time.

10.3   Streaming activities

In 2005, we focused on three major areas:

1. further development of the portal for live streaming
2. further development of the indexing and search portal
3. streaming of high quality video

We also supported the Czech academic and research community in their video streaming efforts and started collaboration with industrial partners interested in a joint development of applications of new technologies and utilisation of our research results.

10.3.1   Development of CESNET Streaming Infrastructure

We extended the announcement portal to comply with the requirements of TERENA TF-VVC. We added localised event filtering, integration with other web servers having a different presentation layer and a number of other minor improvements. The portal was presented at several events from TF-VVC meetings to the DIVERSE conference.

We extended the indexing and search portal with the preview feature that creates several snapshots of the multimedia data in predefined time intervals (currently 0, 30, and 50 s). However, this increases CPU load on indexing distillers and requires increased parallelism in order to achieve efficiency similar to the previous versions. The list of indexed domains grew further and currently includes .cz, .dk, .sk, .hu, .nl, .pl and a part of .edu, representing more than 2,000,000 unique IP addresses. This amount is interesting from the viewpoint of statistical analysis of multimedia formats and metadata information. Compared to the Google or Yahoo! services, we are indexing fewer addresses, but since we are working with metadata, our search results are more relevant.

In the area of high-quality streaming, we focused on two resolutions: PAL and HD. We started collaboration with the MAFRA group and its Ocko station that is oriented on young people and is quite popular among university students. We are broadcasting in two quality levels: the goal of the lower quality broadcast is to experiment with streaming on (i) dual-stack IPv6/IPv4 computers and (ii) PDAs. The higher quality broadcast of 1.5 Mbps is intended as a replacement for the IPTV service. In the future, we would like to broadcast a DV stream (25 Mbps), once we get direct access to its source. Streaming in HD quality is currently limited by the (un)availability of a suitable HD content, as the HDV and especially HD cameras are still rather expensive. We thus decided to build a mobile HDV-based station that could be lent to different groups from the academic community.

The collaboration with Czech Radio continues successfully. We are helping the new science-oriented Leonardo station with transition to digital broadcasting and preparation of live video transmission from their studio in Real Video and Ogg Theora formats. As every year, we participate in interesting projects of Czech Radio - in 2005 it was a project called "Odhaleni - trochu jina reality show" (Revelation - A Slightly Different Reality Show). We handled Ogg Theora streaming, full PAL MPEG-2 transmission for further video processing, transcoding to MPEG-4 format and its delivery via both unicast and multicast.

10.4   Production Infrastructure and Support Services

10.4.1   Streaming Infrastructure

The streaming platform has been expanded by adding a server for streaming in the Ogg format using the Theora codec. As of now, the academic community can work transparently on a uniform platform with the following formats: Real Video, Windows Media, MPEG-1/2/4, QuickTime, and Ogg Theora. With such a number of supported formats, our streaming farm can claim leadership even within the European academic community.

10.4.2   H.323 and SIP Infrastructure

We finished the numbering plan and migrated completely to the 950 08 prefix delegated to us by the IP telephony group. Therefore, we are now fully compatible with all participants of the Vide.Net project. All H.323 components purchased in 2005 are fully operational and integrated into the H.323 infrastructure. Most of the components (see Figure 10.4) also support both H.323 and SIP now.

We created a support toolset for MCU resource control system based on the Polycom XML API. We are distributing a new personal PVX client with mobility support and provide maintenance and software updates for the entire infrastructure and terminals on a routine basis.

10.4.3   Direct Support

Videoconferencing clients are being distributed upon request to CESNET members. We continue testing new equipment as it becomes available on the market and advise our colleagues from CESNET and the Czech academic community about videoconferencing and A/V equipment.

previous

contents

next