October 06, 2011
Last month, the San Diego Supercomputer Center launched what it believes is "the largest academic-based cloud storage system in the U.S." The infrastructure is designed to serve the country's research community and will be available to scientists and engineers from essentially any government agency that needs to archive and share super-sized data sets.
Certainly the need for such a service exists. The modern practice of science is a community activity and the way researchers collaborate is by sharing their data. Before the emergence of cloud, the main way to accomplish that was via emails and sending manuscripts back and forth over the internet. But with the coalescence of some old and new technologies, there are now economically viable ways for sharing really large amounts of data with colleagues.
In the press release describing the storage cloud, SDSC director Michael Norman described it thusly: “We believe that the SDSC Cloud may well revolutionize how data is preserved and shared among researchers, especially massive datasets that are becoming more prevalent in this new era of data-intensive research and computing.” Or as he told us more succinctly, "I think of it as Flickr for scientific data."
It's not just for university academics. Science projects under the DOE, NIH, NASA, and others US agencies are all welcome. Even though the center is underwritten by the NSF, it gets large amounts funding and researchers from all of those organizations. Like most NSF-supported HPC centers today, SDSC is a multi-agency hub.
Norman says that the immediate goal of this project is to support the current tape archive customers at SDSC with something that allows for data sharing. For collaboration, he says, tape archive is probably the worst possible solution. Not only is the I/O bandwidth too low, but with a tape platform, there is always a computer standing between you and your data.
With a disk-based cloud solution, you automatically get higher bandwidth, but more importantly, a web interface for accessing data. Every data file is provided a unique URL, making the information globally accessible from any web client. "It can talk to your iPhone as easily as it can talk to your mainframe," says Norman.
The initial cloud infrastructure consists of 5.5 petabytes of disk capacity linked to servers via a couple of Arista Networks 7508 switches, which provide 10 terabits/second of connectivity. Dell R610 nodes are used for the storage servers, as well as for load balancing and proxy servers. The storage hardware is made up of Supermicro SC847E26 JBODs, with each JBOD housing 45 3TB Seagate disks. All of this infrastructure is housed and maintained at SDSC.
The cloud storage will replace the current tape archive at the center, in this case a StorageTek system that currently holds about a petabyte of user data spread across 30 or 40 projects. Over the next 12 to 18 months, SDSC will migrate the data, along with their customers, over to the cloud and mothball the StorageTek hardware.
According to Norman some of these tape users would like to move other data sets into these archives and the cloud should make that process a lot smoother. "We are setting this up as a sustainable business and hope to have customers who use our cloud simply as preservation environment," he says. For example, they're already talking with a NASA center that is looking to park their mission data somewhere accessible, but in an archive type environment.
The move to a storage cloud was not all locally motivated however. Government agencies like the NSF and NIH began mandating data sharing plans for all research projects. Principal investigators (PIs) can allocate up to 5 percent of their grant funding for data storage, but as it turns out, on a typical five- or six-figure research grant, that's not very much money.
In order for such data sharing to be economically viable to researchers, it basically has to be a cost-plus model. Norman thinks they have achieved that with their pricing model, although admits that "if you asked researchers what would be the right price, it would be zero."
For 100 GB of storage, rates are $3.25/month for University of California (UC) users, 5.66/month for UC affiliates and $7.80/month for customers outside the UC sphere. Users who are looking for a big chunk of storage in excess of 200TB will need to pay for the extra infrastructure, in what the program refers to as their "micro-condo" offering.
The condo pricing scheme is more complex, but is offered to users with really large datasets and for research grants that include storage considerations for proposals and budgeting. And even though this model doesn't provide for a transparently elastic cloud, the condo model at least makes the infrastructure expandable. According to Norman, their cloud is designed to scale up into the hundreds of petabytes realm.
Although data owners pay for capacity, thanks to government-supported science networks , data consumers don't pay for I/O bandwidth. Wide are networks under projects such as CENIC (Corporation for Education Network Initiatives in California), ESNet (Energy Sciences Network), and XSEDE (Extreme Science and Engineering Discovery Environment) are public investments that can be leveraged by SDSC's cloud. That can be a huge advantage over commercial storage clouds like Amazon's Simple Storage Service (S3), where users have to account for data transfer costs.
While some researchers may end up using commercial offerings like Amazon S3, Norman thinks those types of setups generally don't cater to academic types and are certainly not part of most researchers' mindsets. They are also missing the some of the high-performance networking enabled by big 10GbE pipes and low-latency switching at SDSC.
Whether the center's roll-your-own cloud will be able to compete against commercial clouds on a long-term basis remains to be seen. One of the reasons a relatively small organization like SDSC can even build such a beast today is thanks in large part to the availability of cheap commodity hardware and the native expertise at the center to build high-end storage systems from parts.
There is also OpenStack -- an open-source cloud OS that the SDSC is using as the basis of their offering. Besides being essentially free for the taking, the non-proprietary nature of OpenStack also means the center will not be locked into any particular software or hardware vendors down the road.
"With OpenStack going open source, it's now possible for anybody to set up a little cloud business," explains Norman "We're just doing it in an academic environment."
May 16, 2013 |
When it comes to cloud, long distances mean unacceptably high latencies. Researchers from the University of Bonn in Germany examined those latency issues of doing CFD modeling in the cloud by utilizing a common CFD and its utilization in HPC instance types including both CPU and GPU cores of Amazon EC2.
May 10, 2013 |
Australian visual effects company, Animal Logic, is considering a move to the public cloud.
May 10, 2013 |
Program provides cash awards up to $10,000 for the best open-source end-user applications deployed on 100G network.
05/10/2013 | Cleversafe, Cray, DDN, NetApp, & Panasas | From Wall Street to Hollywood, drug discovery to homeland security, companies and organizations of all sizes and stripes are coming face to face with the challenges – and opportunities – afforded by Big Data. Before anyone can utilize these extraordinary data repositories, however, they must first harness and manage their data stores, and do so utilizing technologies that underscore affordability, security, and scalability.
04/02/2012 | AMD | Developers today are just beginning to explore the potential of heterogeneous computing, but the potential for this new paradigm is huge. This brief article reviews how the technology might impact a range of application development areas, including client experiences and cloud-based data management. As platforms like OpenCL continue to evolve, the benefits of heterogeneous computing will become even more accessible. Use this quick article to jump-start your own thinking on heterogeneous computing.