nectar cloud workshop
Originally uploaded by moncur_d.
Just back from the Nectar cloud computing workshop in Melbourne - an event designed to help choose the basic parameters of an academic cloud computing service in Australia.
Now I must admit that before I went to this event I had only a hazy view of cloud computing - to me services like dropbox, skydrive and the various components of the Google ecology represented cloud computing - which of course is really software as a service (SaaS) and it has the capacity to respond to demand and be replicated, so that if one box, or building full of boxes goes down the rest will take up the slack.
On the other hand platform based services such as Amazon's EC2 are slightly different - basically renting computing potential - essentially on demand hosting of virtual machines - so for example if you have a complex application - such as the woodchipper text analysis tool - that you want to run against some data, you would build and load up some VM's - as many as you want, or your credit card will allow to let you do your analysis run in much the same way as my playing with ngram was firing up an instance of the ngram software each time.
Like much in computing nothing is new - only improved - VM/cms with its shared architecture gave you much the same effect on an old IBM or Amdahl mainframe where you effectively had a machine of your own to run your jobs in.
The key differentiators between then and now is (a) the ability to do this with commodity hardware and (b) elasticity - allowing as much or as many on demand. Costing and accounting is of course a complete nightmare.
There is of course an argument as to whether we should bother building a solution in Australia when Amazon and Azure do it better and have more hardware. I've made the argument elsewhere that the future of university computation services lies in mediating access to outsourced services.
However I can see a number of good reasons for building a demonstrator in Australia:
- Offshore data shipping costs - moving a large dataset for analysis outside of Australia is a slow and expensive business - the links, while reasonable, are still expensive and capacity bound when it comes to big data
- There is currently no significant cloud computing provider in Australia - no Amazon or Azure node, the nearest being Singapore, meaning that data shipping costs are a given
- Data security - there is some data - the classic example being health related data - that for legal reasons cannot be shipped out of Australia - and of course there's always the Patriot Act problem when shipping data to the US
- Competence building. Even if the Australian academic cloud service ends up as a front end to commercial providers, we need to know about the services in order to give good support to the client base, and only by playing with this stuff can we build understanding
- Australia is only 21 million people with 39 universities, of which a quarter are research intensive. Individually each research intensive university probably can't build a service to be elastic enough for real use but collectively we just about can
- Having an academic service with a low entry cost lowers the bar for these individual researchers in less research intensive universities who have an occasional need for serious computation - rather than building a beowulf cluster in their office, they can make use of cloud based services for these occasional big analyses - it lowers the bar to their getting on with research