We have been using the Access Grid for over 2 years and yet we have not been able to reliably make multicast connections to other networks, and we have received little support from our network operators to resolve the problems. Discussions taking place on the ag-tech mailing list suggested to me that other people had similar problems, and an inability to use multicast might be widespread in the AG community. Further, a review of the status of the Multicast Beacon also suggests that many sites are having problems making multicast connections.
Widespread problems using multicast would be important for AG users and developers to know. For example, people planning meetings might need to ensure that unicast bridges are always provided. AG developers might also want to place more emphasis on supporting unicast connections, and in documenting the procedures necessary to use unicast.
For these reasons, I decided to poll the AG community to determine how widespread problems with multicast were. I expected to find that most people were not able to use multicast reliably and that there was a frequent need for unicast bridges.
A simple survey was constructed using the free Zoomerang survey service. The survey contained a single question asking people how often they were able to use multicasting for AG sessions, and a second open-ended question designed to allow people to comment about multicast and the AG.
Participants were recruited by messages sent the the ag-tech and ag-users mailing lists. Response were collected for a period of 7 days from Feb 10-17, 2005. A Zoomerang feature of restricting the responses to one response per computer was employed in an attempt to avoid duplicate responses.
The survey pages received 120 page visits during the response period, and 91 people completed the survey.
The quantitative section of the survey contained a single question:
Participants were asked to choose among 5 responses, or to select an "other" response and provide a comment. The response choices, along with the frequency of responses and the percentage, are shown in Table 1.
| Response | Count | Percentage |
| 1. I am never able to use multicast | 10 | 11 |
| 2. I usually cannot use multicast | 7 | 8 |
| 3. I use multicast and unicast equally often | 4 | 4 |
| 4. I usually can use multicast | 33 | 36 |
| 5. I am always able to use multicast | 33 | 36 |
| Other, Please Specify | 4 | 4 |
An analysis of the answers from the 4 people who chose the "other" response showed that:
The final results after the recoding of the "other" responses are presented in Table 2.
| Response | Count | Percentage |
| 1. I am never able to use multicast | 11 | 12 |
| 2. I usually cannot use multicast | 7 | 8 |
| 3. I use multicast and unicast equally often | 4 | 4 |
| 4. I usually can use multicast | 34 | 37 |
| 5. I am always able to use multicast | 35 | 38 |
The results in Table 2 show that the majority of responders (75%) are able to use multicast most or all of the time. Only 20% of the responders reported usually or always having problems with multicast. These results suggest that problems with multicast are not as common as I expected.
The qualitative section of the survey contained a single question:
A total of 50 responses were collected. Using the approach described in Grounded Theory, the comments were coded by making notes of important statements or themes in the text. The notations were then summarized until the essential concepts emerged.
The most interesting concept to emerge in the qualitative analysis is a striking difference in the approach to unicasting. Two responders emphatically described unicasting as a crutch that should be avoided: "using crutches like a unicast bridge should only be done as necessary to convince upper management of the value of AccessGrid" and "Unicast bridges are a convenient workaround, enabling network engineers to not do their job." In contrast, 4 people described unicasting as being essential for the success of the AG. For example, one person wrote "unicast bridging is critical to the success of a collaboration tool such as AG." In addition, another 4 people took a more pragmatic approach and describing unicasting as a valuable backup method to be used when multicasting is not available. An example of this type of response was: "most of the venues that we visit are bridged for the occasion when [multicast] fails."
Another theme in the comments, mentioned 6 times, was lack of institutional support for multicasting. A typical comment in this theme was: "our host institution resists allowing widespread usage of multicast." Perhaps a related theme was that many multicast problems arise at the local level, often at the scale of the individual LAN (mentioned 4 times). For example, one participant wrote: "Multicast on the WAN is ok, on our Lan is a problem." When describing the problems that they have, two people described experiencing one-way multicasting, where traffic flows in one direction but not the other.
Another emergent theme in the comments was problems debugging multicasting and a lack of diagnostic tools (mentioned 5 times). A typical comment in this theme was: "am not clear on how to troubleshoot multicast."
Some people also described multicasting as unreliable (mentioned 4 times), with a typical comment being: "multicast has been totally unreliable, with two years experience." At the same time, people sometimes described multicast connections as being better than unicasting because the tools worked better or bandwidth is used more efficiently (mentioned 4 times). One example of this type of comment was: "the media tools launch and run better when using multicast."
Three of the responders stated that multicast worked for them, but other people that they interacted with often had multicast problems. Specific problems were also reported when using multicast for sessions that connect overseas (mentioned 3 times), with an emphasis on problematic connections between the UK and the USA.
Contrary to my expectations, the majority of people reported that they
are able to use multicasting for their AG sessions some or all of the
time. The results do suggest, however, that multicasting is not without
problems and unicasting has an important role today, both for the
minority of
people who do not have any multicasting ability, and for other people
who rely on unicasting on the occasions when multicasting fails. It
might be interesting to pursue the comments made by two of the
responders that suggested that unicasting is a crutch that should be
avoided, since this seems to be contrary to many of the other opinions
and the current use of unicasting.
The reported problems in getting institutional support for multicasting is a continuing concern. It is likely that these problems are related to some of the other issues that were identified, such as the lack of debugging knowledge and tools, and problems associated with LANs. It is possible that more information and tools directed at decision makers and network operators in institutions might be helpful.
As these results are interpreted and discussed, potential problems with sampling should also be mentioned. This was a voluntary survey directed at two AG mailing lists. It is possible that people who use multicasting successfully had a higher tendency to participate and the responders are not representative of the total AG community. It might be that other AG users are having more problems using multicast but they did not receive the recruiting messages or chose not to respond. To be sure, a wider, more systematic survey would have to be conducted, and it would be best if independent collaborating evidence of the general success of multicast was also available.