Conventional wisdom, if there is such a thing, suggests that the rise of cloud computing will bring a resurgence of thin-client technology back to the corporate workplace. While this is true, the cloud may also be the harbinger of the workstation–high-end personal computers that, when coupled with the cloud, could bring more disruption to the supercomputing sector.
This seems counterintuitive, since the promise of cloud computing was to shift computational workload from the client in the client-server model back to the servers. The end user only needs a browser to access web-based apps, the reasoning continues, so why give them anything more than a thin-client with a big network pipe?
In many situations, this is indeed the case. There is even a trending extreme of this line of thought: zero client machines provide appliance-like functionality while absorbing little power and ease-of-use and -maintenance, thanks to centralized over-the-wire administration.
But increasingly, IT managers are seeing a trend in the other direction: more powerful PCs that can easily be classified as workstations. Bigger CPUs, faster graphics, and upwards of 20 Gb of RAM–these are machines that are designed to do a lot of local processing.
The idea, it seems, is to take advantage of cloud technology and economics and use workstations to bring high-performance computing (HPC) capabilities to the corporate user or any user connected to the Internet with a big enough connection. To accomplish this, cloud vendors are working with applications that live and do some work on the local workstation, but then shunt much of the workload out into the cloud and back again. This approach enables users to get HPC-level power with little to no special configuration or hardware beyond the workstation itself.
InterGrid’s green button
One such vendor with this approach is InterGrid and its GreenButton software, which is working with independent software vendors to build the apps for this kind of architecture. The New Zealand-based company teams up with existing ISVs and their projects and gets those projects “GreenButton enabled.” A user can install one of these GreenButton-enabled applications on their workstation and then get HPC using Microsoft’s cloud services platform Azure.
A variety of applications currently make use of GreenButton functionality, mostly centered across the high-end graphics applications that would need hours of processing time to render locally, but may only need a few minutes when tapped into GreenButton’s service. Despite GreenButton’s reliance on Azure, Linux users should note that GreenButton can be attached to Blender and Maya.
What both workstation-in-the-cloud technologies offer is the capability to increase the number of resource-intensive jobs that can be carried out in parallel. There is still a need to have locally heavy power and graphics, but when it comes to graphics rendering, simulations, or biometric processing, the parallel processing of the cloud can be leveraged to minimize job times.
If this sounds a bit like grid computing, you would be close, but the key difference here is that unlike grid, plugging workstations into the cloud doesn’t require the administrative overhead of managing a grid computing infrastructure. Local applications and operating systems will still need to be configured, of course, but the burden is still less than managing a grid.
This approach can be less painful economically, too. GreenButton, for instance, offers a $19.95 base monthly subscription rate for 100 allocated cores, 100 maximum users, but only one job at a time, or a $0.30/core hour rate with 360 cores/pool, six users, and the capability to run simultaneous jobs.
Other services will offer similar rates, likely tied into the rate plans from Azure and Amazon Web Services.
Disruptions ahead for supercomputing
Once such services become more prolific and more ISVs provide the apps to work with such platforms, then it will be only natural to see even more disruption of supercomputing, particularly as verticals like geological research and biotechnology start to make use of such systems in favor of high-cost local supercomputers, which are inherently limited by their size. Plugging into cloud-based processing, however, isn’t limited by hardware: the only limit is your budget.
There are trade-offs, of course: though administrating this kind of architecture costs less than grid computing, it’s still about the same as managing a typical client-side architecture. And let’s not forget, workstations aren’t cheap. The average desktop PC runs for about $500, while workstations can average around $1100. These costs will definitely have to be considered if you plan to deploy this kind of architecture.
Still, now that customers can have direct access to scalable supercomputing power from within native applications with which end users are already familiar, the possibility of getting months of lead time on competition thanks to much-shortened simulation or processing times.
That kind of benefit could be the difference between winning and losing in today’s tightly competitive markets.