Riddle me this…

Arn Hyndman October 3, 2008 in: Featured Posts

A while back, Christian Renaud posted an interesting question “Riddle me this…” on his blog:

Fast forward to the office of the future, the virtual office version.  You have packetized spatial audio.  You have user created content.  You have streaming video and powerpoints and presence information.  You have ever-changing mixes of synchronous and asynchronous traffic types all over walls and tables of your virtual headquarters.  This is much more bandwidth intensive than Warcraft, and if you are having a staff or funding meeting, the voice/video latency is arguably more critical than simple ‘the dragon killed you before you hit it with your sword’ telemetry data […]

So, given that you can’t ’shard off’ the different offices (you are trying to facilitate the ‘death of distance’, right?) and you need to have a single virtual office where people can intermingle and interact richly, how are you going to accomodate the insane bandwidth and latency requirements of a fully-annotated virtual environment? Or, are the geographically remote participants going to have to suffer the latency inherent in such a rich environment as the cost of global collaboration?

This is going to be the multi-million dollar question as virtual worlds continue to mature. I think that at least part of the answer is going to be the same as those developed for existing web media delivery. That is, cache the media close to the consumer. This solution works well for web pages, graphics and streaming video; the same solution will certainly work just as well for delivering the relatively static 3D content and media that makes up the bulk of the data required to render a virtual world.

This leaves the problem of dynamic user created content (3D content that must be visible to other users as it changes in near real time), application sharing, avatar telemetry, text messaging and voice. Of these, I would suggest that dynamic user created content, avatar telemetry, and text messaging are relatively low bandwidth and thus easier to solve. The challenge will be with application sharing and voice.

For both application sharing and voice, many current solutions use peer-to-peer communications. This architecture has several benefits: (1) no network or cpu cost for the virtual world host (2) good performance for small numbers of peers that are near each other in network topology sense. Unfortunately, peer-to-peer solutions don’t scale well to large numbers of users. In the worst case, n users communicating to each other results in n x n streams – naturally, the bursty nature of voice and application sharing allows optimizations that generally avoid this worst case scenario, but regardless, simple peer-to-peer communications present significant scaling problems which I believe will lead to their eventual elimination from use in enterprise virtual worlds.

It will be interesting to see how solutions evolve over the next few years to handle the problem of scaling voice and application sharing applications for global enterprise virtual worlds.