HPC | News

Cornell Launches Private Cloud for Research

• By Dian Schaffhauser
• 11/01/11

Cornell University has launched an on-demand research computing service available to scientists inside and outside of the institution. Red Cloud, named after Cornell's school color, is hosted by its Center for Advanced Computing and brings online new equipment.

The private cloud research service comes in two subscription flavors. The basic offering, called "Red Cloud," is an Infrastructure as a Service (IaaS) that runs an open source version of Eucalyptus, a cloud computing platform. Subscribers gain root access to virtual servers and virtual disks.

The second offering, "Red Cloud with MATLAB," delivers MATLAB Distributed Computing Server as software as a service. This edition of MathWorks' numerical computing environment lets users solve complex problems by executing MATLAB and Simulink on a computer cluster. Users program and prototype applications on their desktops using Parallel Computing Toolbox and then scale up to Red Cloud using MATLAB Distributed Computing Server. The Red Cloud environment for MATLAB includes Nvidia graphics processing units, which improves processing performance for these types of problems.

According to Center Director David Lifka, Red Cloud was developed to address two scenarios that weren't being covered well at Cornell. One case is the user who needs a tiny bit of computing power for a long stretch of time, "for example, someone who wants to run a research wiki to collaborate with other researchers, which doesn't need hundreds of cores and tons of memory. It needs one core and a little bit of memory. They don't want to go out and buy a server for that."

The other case is the user who needs "lots of cores," but only for a short period. "Maybe I have a big paper due and then I'm done, and I don't want to buy a machine to have it sit idle for the rest of the year," he explained.

A single subscription to Red Cloud provides 8,585 core hours, roughly equivalent to the use of a single core for about a year. A user can choose to churn through that usage more quickly by processing with more cores simultaneously. The Red Cloud subscriptions provide 10 gigabit Ethernet data connectivity in and out of the cloud. Expert consulting is also available by Cornell personnel.

The pricing for a single subscription by Cornell units is $500 for the IaaS edition and $750 for the MATLAB edition. Other academic institutions can subscribe for $750 and $1,200, respectively.

That subscription model is important to the Center's users, Lifka noted. "Researchers don't like things that are pay-as-you-go. Why? Because there's this deep-seated fear that their grad students will get on some weekend and run up some big bill--like giving your kid a cell phone and having them think they have unlimited texting, and they don't." By tying usage to a limited subscription, their risk is bounded, he explained. "If your grad student goes nuts, we're going to stop them at that number. If you want to add more time, you can. And they like that."

Subscribers can choose from among five virtual server setups, from a minimum configuration of a single core with four GB of RAM and 20 GB of disk space to a maximum configuration of 12 cores, 48 GB, and 1 TB of disk space.