With ISC19 fast approaching (June 16-20) ThinkParQ is ramping efforts to showcase its BeeGFS parallel file system’s growing market traction and to draw attention to the company’s efforts to develop the needed ecosystem of technology and channel partners to make BeeGFS successful. In a sense, it’s starting to feel like 2019 is BeeGFS’s break out year or at least its coming out year.
Today, ThinkParQ and NVMe block storage specialist Excelero released benchmarks in which Excelero’s NVMesh product helped BeeGFS achieve substantial speedups in file creation, metadata read, and IOPs. At ISC, ThinkParQ will have a range of activities including a BOF, which among other things will discuss forthcoming technical feature plans; host a joint demo with NetApp at ThinkparQ’s booth; and tout its growing channel.
BeeGFS, you may know, got its start as an in-house project (2005) at the Fraunhofer Institute for Industrial Mathematics (ITWM) and was called the Fraunhofer Gesellschaft File System (FhGFS). It was spun out in 2014 as BeeGFS under ThinkParQ and the race to commercialization began in earnest. While BeeGFS is “open source,” there are proprietary elements and services sold by ThinkParQ.
Helped by the relative turmoil in the parallel file system market – Lustre ownership changes (Intel to DDN) and IBM’s rebranding of GPFS as Spectrum Scale – BeeGFS has been steadily gaining ground including building up a channel. (For more background on BeeGFS, see HPCwire article, Move Over Lustre & Spectrum Scale – Here Comes BeeGFS?)
Here are four milestones in the last couple of years that highlight BeeGFS gains outside of Europe:
- CSIRO. Dell EMC has been working to deploy BeeGFS at CSIRO (Commonwealth Scientific and Industrial Research Organization) in Australia.
- TACC. The Texas Advanced Computing Center has publicly said it’s evaluating BeeGFS and two national laboratories are reportedly doing the same.
- Inspur and Pacific Teck. Inspur has become ThinkParQ’s exclusive distributor for China while Pacific Teck is leading a broader AsiaPac effort.
- ABCI. Japan’s AI Bridging Cloud Infrastructure has deployed BeeGFS.
Today’s announcement was intended to further highlight BeeGFS’s NVMe-friendly nature at a time when NVMe is hot and especially useful for accelerated systems often slowed by IO bottlenecks. The CSIRO installation, for example, is a two-petabyte all NVMe solution targeting AI. Working with Excelero extends BeeGFS’s reach and enhances both companies’ go-to-market capabilities. Excelero offers software defined NVMe block storage. The companies have already successfully deployed NVMe storage clusters with BeeGFS and NVMesh to increase productivity of advanced GPU compute clusters.
Here’s an excerpt from the announcement:
“To demonstrate the possibilities of this new scale-out infrastructure, the companies used the industry standard mdtest and IOR benchmarks. The test system, say the companies, was a compact 2U 4-server chassis with a total of 24 NVMe drives, connected via a 100Gbit RDMA network to 8 BeeGFS client compute nodes. Tests were run on the exact same hardware configuration with BeeGFS utilizing the direct-attached NVMe vs. BeeGFS utilizing NVMesh logical volumes.
“Taking advantage of NVMesh to offload mirroring operations, BeeGFS file create operations were boosted 3x, while metadata read operations were boosted 2.5x. For small random file access, which is often considered to be especially critical for application efficiency, NVMesh’s low latency technology boosted BeeGFS 4K write IOPS to 1.25 million per second, a 2.5x improvement. By leveraging NVMesh distributed erasure coding for BeeGFS, customers can get up to 90% usable capacity while still tolerating drive failures. All the while, achieving 75GB/s streaming throughput from this entry-level system,” according to the release. (HPCwire has a few more details on the new offering here.)
ThinkParQ CEO Frank Herold provided HPCwire with a progress update prior to today’s announcement and while he touched on the Excelero collaboration he focused on broader plans for ThinkParQ’s commercial growth. Adding channel partners been a priority as has been establishing a larger footprint outside Europe. In 2019 alone, the company struck deals with Pacific Teck (February, AsiaPac), Inspur (March, China), and E8 Storage (April, U.S. and Israel) for varying types of collaboration. Last summer U.S.-based Advanced HPC became a “Platinum Partner” partner. Recently, ThinkParQ realigned its fiscal year with the calendar year. After a modest presence at SC18, the company will have a booth this year. As noted earlier, a wide range of activities is scheduled for ISC.
Landing Inspur was an important win. “We had our first orders from Inspur and that [seems] to be working well. China is hard to enter as a remote company. We are selling maintenance contracts only (Inspur) so it is not really selling the software stack,” said Herold. He noted it’s a price sensitive market with substantial regulatory issues and that Inspur is better able to handle those challenges.
The Pacific Teck deal is aimed more broadly at Asia. “I would not call them a distributor but would call them a go-to partner for BeeGFS in the region,” he said adding India is a key target with a good pipeline of “large opportunities” coming up.
The U.S. is of course a large market. Given the CSIRO project with Dell EMC, it’s natural to wonder if a more formal partnership might be in the works. They are already working together with a few customers, Herold allows, but declined to say more. Likewise, ThinkParQ is working with a large U.S. automotive manufacturer and a significant financial institution in the U.K. “We’re hoping to be able to talk about the automotive customer [reasonably soon],” he said, noting that financial services customers rarely talk publicly about what they are doing.
Noteworthy, there are several emerging options in the storage technology space. WekaIO is one that has had early success. Its marketing pitch is: “Matrix software is an NVMe-native, resilient file system that runs on commodity servers, delivering the highest bandwidth, lowest latency performance to any InfiniBand or Ethernet enabled GPU or CPU based cluster. Integrated tiering automatically and transparently migrates your cold data to any S3 or Swift compatible private cloud for low cost and limitless scale.”
Said Herold, “WekaIO has a completely different approach and go to market model, and different pricing. It’s a good product. I would say the target customer they are looking for is potentially different than [ours although] there is some overlap. We are focusing on customers at larger scale in mixed environment having NVMe but also having traditional storage needs. Weka is mainly focused on the NVMe side [and] expensive compared to what we offer.”
As always the market will be the ultimate arbiter.