Whamcloud to Put New Sheen on Lustre
Aug 24, 2010 5:00 AM PT
The founders of Lustre software technology company Whamcloud opened for business in June with a lot of potential and years of experience working with high-performance computing. What they lacked from their first day was any signed contracts.
Whamcloud Cofounder and CEO Brent Gorda still is waiting to sign the dotted line with his company's first customer. Given the popularity of the open source Lustre project, he is sure that will only take a bit longer to achieve.
Having a commercial company serving as a front office for Lustre technology is viewed as a mighty big deal by the Lustre community. Lustre is a parallel file system that runs across clusters of storage servers to provide the high data bandwidth that supercomputers require. It was developed through funding from the U.S. Department of Energy.
"We have potential customers vying for that first announcement," Gorda told LinuxInsider.
What It Does
Whamcloud added refinements to the Lustre project software. Since the supercomputing field is a niche market, little specialized product support outside the community itself is available to users. That is where Whamcloud steps in.
Its development of Lustre is producing a software product that runs significantly faster in the cloud and does so in a manner that is much closer to the look and feel of an NFS file system, according to Gorda. So you get a file system that can operate at dramatically faster speeds and resiliency in the face of error.
Supercomputing hardware runs 10,000 or 20,000 disk drives all day. So you can image the problem if you keep losing data from the drives that fail, he explained.
The software makes sense of all of that and also allows the system the aggregate performance that those disks can provide to the system. On other computing platforms, data is saved to a disk nearby or at a remote server. Lustre software spreads that file across 5,000 or 10,000 disks, depending on a lot of different factors.
Full Speed Ahead
"When all of those disks operate in parallel, that's where you get the speed. You need the resiliency because at any time one or more of those disks could be dropping out. So that is a key thing," Gorda noted about how Whamcloud is improving Lustre.
In the high-performance community there are feature requests to increase the performance or do it over a wide area. To meet that functionality, Whamcloud is conducting tests over the next few months involving storing data at 100 GB per second on a network, Gorda explained.
"That is dramatically faster than any of the standard storage that people can do now," he said.
In the world of supercomputing, one size definitely does not fit all. That may be a condition that Whamcloud is planning to challenge.
The Lustre technology came out of the U.S. National Laboratories. It is currently one of two competing products to handle the parallel file system used in high-performance computing. It is the only open source project. IBM developed a proprietary file system.
This is a fragmented market. Not one of the parallel file systems fully meets the needs of all users. Cluster systems are growing so fast in size that it's difficult for any one system to keep pace, Steve Conway, research vice president for high performance computing at research firm IDC, told LinuxInsider.
"Lustre is a real Cinderella story," said Gorda. He sees his company as producing the magic slippers.
One of the things that makes Whamcloud's version of the Lustre software unique is that it is neutral with respect to hardware. So all of the leading hardware will run this software. Whamcloud will help its eventual customers maintain it.
"We have a hardware-agnostic solution. It is definitely Lustre on the Linux operating system. But the hardware behind it -- the RAID controllers and the disks the enclosures and all that -- are completely agnostic," he said.
Jumping into a business with only potential customers in tow is risky, but given the niche high-performance computing market, Whamcloud hopes to offer what nobody else is providing. Plus, the Lustre community is strongly behind the effort. Whamcloud's founders have a long history with development of Lustre technology in its formative years.
Peter Braam was a researcher at Carnegie Mellon University who founded the Lustre Project [*Correction - Aug. 24, 2010]. He also founded Cluster File Systems (CFS) to provide commercial support for the parallel file system in 2001.
Some five years later, Sun Microsystems bought CFS for its HPC business and merged it with its own Zettabyte File System (ZFS). Now Oracle owns the project as a result of buying Sun in January. Brent Gorda previously worked on the development of Luster at the U.S. DoE.
Plus, Gorda worked for many years on Lustre at National Laboratory. Eric Barton is Whamcloud's chief technology officer. He worked at Lawrence Livermore National Laboratory and later became a principal engineer for Lustre after Sun bought CFS.
"We believe in this technology and in high-performance computing," said Gorda.
The birth of the company resulted from a love affair with the Lustre Project community and a sweatheart deal with investors. The community felt that support for Lustre and its continued development had to be in the hands of a serious company. So community leaders lobbied Gorda and Eric Barton to step out of their comfortable roles at National Labs and Oracle and really put their money where their mouths were, said Gorda.
"We took [US]$10 million, which we believe is a nice amount to keep us living from hand to mouth. But we also have significant community interests. The community of high-performance people come from government labs around the world as well as large businesses," Gorda said.
Whamcloud's founders found an open-arms reception. That has been extremely encouraging, he added.
Polishing Lustre's Shine
Whamcloud's goal is to push the technology forward to the next level of performance in high-performance computing. The hardest part: establishing mindshare. Each time Gorda talks to a group of people about Whamcloud and the Lustre technology, each listener hears something different.
For example, Gorda said he had recently spoken to a group of potential customers. Then, a short time later at a technology conference those same would-be customers announced that Whamcloud was already in the process of being acquired.
"So I'm finding that I have to spend a lot of my time just talking to people and trying to show them that we are taking our time and trying to do this in a controlled manner. It takes a lot to explain that we want to become a strong player in the community," he lamented.
Beyond Customer One
Whamcloud is focusing in the near term on the use of Lustre technology as a speedster component for high-performance computing. That technology is capable of controlling tens of thousands of hard drives.
The speed factor could become a hard-to-resist selling point for Whamcloud. The latest lab test results showed file saving and access speeds that surpass 240GB per second running on some 20,000 hard drives simultaneously, according to Gorda.
The supercomputing file system is a critical part of the work flow. The data volume has exceeded local storage capability.
"That means you no longer procure hard drives along with your super computer because then you have to store your data locally, impeding access to it from all users. Instead, data is stored on the network," said Gorda.
The speed of Lustre's performance is not the only thing accelerated with Whamcloud. The high-performance computing field itself has grown fast. Around the year 2000, Lustre became more valuable as the market started using more common hardware, according to Conway.
"The market for supercomputing started growing at 20 percent per year starting around 2002. That $2 billion market grew into $8.6 billion today just for the hardware," Conway said.
Clusters became the dominant species in high-performance computing. Now hardware fills up huge rooms with multiple systems hooked together with lots of users wanting to share files, he said.
That demand for shared storage will open the door to cloud computing for HPC users, noted Gorda. The current thinking is that merging cloud technology with HPC will create a synergy that will be very beneficial to both the Lustre technology and cloud computing, he explained.
"There is an overlap of technical requirements that is very interesting to us," he said.
Room for One More
Lustre technology is largely a Sun Microsystems phenomenon now. But it is also an open source entity, much like a Linux developer, Conway said.
"There is definitely room for other vendors coming in with commercial versions of Lustre. It does not belong wholly to Oracle. It is a fragmented market. Lustre is improving, but it has a long way to go," he said. "Anybody with the gumption to improve Lustre will be well positioned to participate in the anticipated growth of the market."
That is exactly what Gorda plans to do through Whamcloud. Right now, all that is standing in this startup's way is signing its first paying customer.
*ECT News Network editor's note - Aug. 24, 2010: The original publication of this article incorrectly referred to Peter Braam as a Whamcloud cofounder. He is not an actual cofounder of the company.