Due to the size of datasets to be crunched by supercomputers, cloud services offering such abilities won't get off the ground without the National Broadband Network (NBN), according to high performance computing specialist Xenon.
"What we're finding in Australia is that the cloud without NBN is pretty much a non-event," Xenon managing director Dragan Dimitrovici said yesterday at a briefing on graphics processing unit (GPU) computing.
He said the idea of the cloud, with everyone having their own private facility that they expand out seamlessly, was hindered by a lack of bandwidth in the current Australian market.
To battle this, Xenon has been renting GPU clusters and then moving them on location to the customer. That customer can do simulations for a couple of months with the clusters, then pass the kit back.
CSIRO post-doctoral fellow Dr Luke Domanski also said that bandwidth was an issue with cloud.
"Traditionally, sciences have worked with much larger datasets," he said. "You do have that bandwidth issue where if you want to run something remotely, you've got to get the data there."
He wasn't sure, however, that this issue could be fought via the NBN.
"I think every time you make something faster, people are going to find something bigger," he said.
CSIRO bought a GPU high performance computing cluster in 2009, taking the plunge into GPU because "multi-core is the future", according to Domanski. Many research projects have been able to increase the speed of calculations, allowing them to carry out simulations that were temporally unfeasible previously.
The research centre has also invested in a number of supercomputer desktops, which are multiple-GPU units that deliver teraflop grunt for a scientist's use.
Nvidia GM Tesla computing Andy Keane said that there would only be two types of use for high-performance cloud computing.
The first was for very peak usage, such as that seen for movie production houses, where there would be a sudden need for a lot of power, then no need at all. The other was for development, for instance in the financial services sector.
In other cases, not only would it be difficult to transfer large datasets, which he said could be as large as petabytes in the oil industry, but there was no desire to take those valuable datasets from the safety of the organisations.
Keane also pointed out that if organisations were using their high performance computing clusters 24/7, they would likely get a better return on investment if they bought them.
However, Keane believed that the greatest barrier to cloud wasn't the size or sensitivity of the datasets, but rather that university grants were geared towards making capital investments, not spending money on operational cloud expenses.