It seems to be an article of faith these days that Cloud Computing is the way to go for cheaper, more connected computing services, but a recent spate of outages at Amazon, Google and other providers has raised some questions about over-reliance on centralised services and revived interest in solutions that live on the network edges such as Peer-to-Peer (P2P) services.

Whilst the outages may knock some people's confidence, cloud computing remains more reliable and significantly cheaper than just about any internal IT infrastructure. But far more worrying are the issues associated with dependency on centralised storage and control. Boing Boing covered the story of somebody whose entire Google identity and hosted data disappeared without trace. Extremely painful for an individual, but potentially lethal for a company.

Right now, our company delicious account appears to have disappeared, and the support response from the Yahoo-owned service has so far been MS-paperclip poor ("I understand that you need to retrieve your username or password"). Luckily, we have a local copy of all our bookmarks going back to 2005, so we can rebuild and move to another service; but if we didn't, then that would really be a blow. At this point, I have no idea why the account has disappeared, and it may not even be Yahoo's fault, but the impact of over-reliance on externally-hosted data is the same whatever the cause.

Over at Read-Write Web, Gmail downtime prompted speculation about a return to P2P rather than centralised cloud infrastructure. They also looked at Wuala, a P2P/Cloud service for data storage, but suggested that this might not be a solution for business because of the unknowns around where your data is physically stored in a P2P network. And for me, 'where your data is stored' remains an issue. Given the political situation in some countries (e.g. the United States), holding data in cloud services physically based there means risking forfeit or intrusion by unpredictable government bodies. Amazon offers a European storage option in additional to the USA, which is a useful option.

There is clearly a lot of potential for businesses to utilise cloud computing to reduce IT costs and increase reliability, but I think we will see the emergence of hybrid models where (high-redundancy, locally mirrored) cloud systems deal with non-secret, non-critical data, but combine with local storage for risk-averse data. Perhaps there is also space for a decentralised P2P cloud network that does not depend upon a single company and is not subject to a single legal jurisdiction. Thinking about our own usage, we could probably use P2P storage and backup facilities as long as we were relatively happy with the encryption used. It certainly has the advantage of redundancy over centralised cloud services. I am sure we will continue to use the cloud for commodity services and storage as well. But I suspect that even here there is a certain class of data that will remain locally stored and inaccessible to the cloud for a long time to come.

The P2P cloud notion is an interesting one, and much closer to the spirit of network intelligence and decentralisation than the centralised cloud services most of us use today. But why stop at storage and services - how about net access itself? Danny O'Brien recently suggested that P2P might be worth looking at for cheaper and more robust city networks:

The cost of providing high-speed Internet to every home in San Francisco is over $200 million, the study estimates. But what is the cost of one person or business making a high-speed point-to-point wireless connection to a nearby Internet POP, and then sharing it among their neighbourhood? Or even tentatively rolling out fiber from a POP, one street at a time? I suspect many people and businesses, don't want HDTV channels, don't want local telephony, and don't want to wait ten years for a city-wide fiber network rolled out: they just want a fast cable on their end, with the other end of the cable plugged into the same rack as their servers.

I think Danny is developing a nice line in think pieces about distributed network infrastructure, so look out for his talk at next year's SWSX, assuming it gets selected.