During the past decade, parallel file systems have redefined how we think about I/O performance. The largest parallel file system storage environments today deliver TB/sec of bandwidth while, at the same time, sustaining hundreds of thousands of metadata operations.
As a company, DDN® has grown and prospered together with Lustre*, which to this day remains the dominant open source parallel file system for HPC and technical computing.
The first large-scale Lustre system fully implemented and supported by DDN was at the National Center for Supercomputing Applications (NCSA) in 2003. That 1,280-node tungsten cluster had 104 OSS connected to 140TB of storage, delivered greater than 11 GB/sec of sustained I/O speed, and ranked number 4 on the Top500 list.
Today, DDN maintains and supports what is by far the largest and most diverse Lustre and file-system user base in the industry. Launched in 2009, DDN’s EXAScaler® has effectively become the industry’s de facto Lustre distribution standard for the Technical Computing, Big Data, and AI markets.
Our largest Lustre customers today run production environments with hundreds of OSS servers, well over a hundred petabytes of storage, and several terabytes/sec of I/O performance. Interestingly, DDN’s number one Lustre customer is not a research organization or DoE lab; it is a for-profit corporation with extremely strict uptime and support requirements.
To provide our end users with the best Lustre experience bar none, we have built a highly sophisticated test environment with hundreds of automated tests running in parallel, and we spend many months hardening community Lustre versions to enhance stability, uptime, and performance.
Better yet, DDN’s EXAScaler comes with a time-proven and tested approach to high availability in Lustre, and with an extensive set of tools that simplify deployment, maintenance, and support.
Future versions of DDN’s EXAScaler currently under development contain important DDN-developed capabilities to Lustre, such as a powerful new project quota feature for ldiskfs-based Lustre systems, significant performance enhancements for a broad range of I/O scenarios, a far stronger policy engine design, and a sophisticated approach to Lustre monitoring. And with DDN’s Infinite Memory Engine® (IME®), we have created a groundbreaking, transparent, and elastic SSD-caching layer, which is fully integrated with the Lustre file system.
Importantly, unlike some of our competitors, DDN has never forked the community Lustre tree, and we have consistently provided enhancements, feature developments, and bug reports back to the community.
In the years to come, DDN will continue to work very closely with the Lustre community and with the core Lustre development team at Intel. We are significantly stepping up our investments in Lustre, continuing to add new features, enhancing Lustre reliability, resiliency, and uptime, expanding our internal development and support capabilities, and further improving our automated test and QA infrastructure.
The net result of DDN’s increased investments in Lustre is simply to provide our diverse customer base with the most compelling, richest, and best possible Lustre experience in demanding production environments.
Please feel free to contact me directly with any outstanding questions.
Questions and Answers
Q: I understand that, under the hood, DDN’s EXAScaler uses the Intel Enterprise Edition of Lustre (IEEL). Intel has effectively announced that it will consolidate all of its efforts to a single-community Lustre release. How does this affect EXAScaler?
A: DDN’s EXAScaler is a greatly improved Lustre distribution that has been built, enhanced, and hardened by DDN. In order to provide customers with the best Lustre experience, we have built EXAScaler on top of the most advanced and most tested version of Lustre rather than forking the code basis, as some of our competitors have done. During the past few years, we have used Intel IEEL as a starting point for EXAScaler. In the future, we plan to use the Intel Long-Term Stable (LTS) community release, and thanks to our increased investments and staffing in Lustre-related activities, we will continue to add features, capabilities, and reliability to enhance our customers’ Lustre experience.
Q: Is there any impact of the Intel announcement on existing EXAScaler versions or EXAScaler support contracts?
A: Absolutely not. There is no impact on existing customers or contracts whatsoever. Our relationship with the HPDD Team at Intel remains strong and Intel will continue to provide additional backend support to DDN Technical Teams.
Q: How will the discontinuation of IML (Intel manager for Lustre) affect DDN?
A: EXAScaler uses DDN’s own high-availability stack and management tools. Unlike our competitors, we have never used IML for management.
Q: Will DDN continue to work with Intel on Lustre?
A: Absolutely. Software is developed by people, not corporations. And the foundation of complex, system-level software, such as a file system, tends to be developed by a small group of people. Today, the core Lustre development team is at Intel. Through the years, we have worked with this team at various companies, from CFS, to Sun Microsystems, to Whamcloud, and most recently Intel. We will continue to work with the core Lustre team, wherever that team is based, and we will supplement their efforts with our own expanding DDN Lustre team. Intel is an important strategic partner for DDN, well beyond Lustre, and we will continue to work closely with the Intel teams that develop platforms, interconnects, and storage devices.
Q: The contract with my existing Lustre support provider is expiring, but I am not using DDN hardware.
A: Absolutely yes. We actually have provided Lustre and EXAScaler as “software only” for a number of years. Also, we do provide a broad range of consulting services to upgrade your Lustre file system to the most stable version for your environment, or help you to optimize Lustre or EXAScaler deployments on non-DDN hardware. However, as a storage vendor, we have always taken a system-level perspective toward Lustre. The stability of any Lustre or EXAScaler system critically depends upon its hardware components. Thus, the SLAs we provide for “software only” Lustre or EXAScaler support will differ from the SLAs for a complete EXAScaler solution.
Q: Will DDN provide support for Lustre or EXAScaler with ZFS?
A: DDN has worked for quite some time on a ZFS-based solution, and we expect to deploy our first ZFS systems during the next few months. But, we feel it is important that our end users understand what ZFS truly means. The most important goal for a storage solution is to protect data. In the case of ZFS, data redundancy is provided by the open source ZFS file system, which is supported by a very small group of programmers in the open source community. Thus, in the case of a catastrophic failure, data recovery can be challenging, time consuming, and in some cases impossible. It is important that our end users understand when ZFS is an appropriate option, and when not. The good news is that DDN has a team of ZFS experts on board, and we can help our customers make the best choice possible for their needs.
Q: I understand DDN has hired some key personnel from the former Intel High-Performance Data Division (HPDD)?
A: It is true that a number of key members of the Intel HDPP division now work at DDN. We always strive to hire the best talent available in the industry.