Enterprises need unified NFS+SMB file services, and files and file servers will outlive us all
retained text
FILE Enterprises need unified NFS+SMB file services, and files and file servers will outlive us all Interview: B&F was contacted by Ned Pyler, CTO Enterprise Storage at Tuxera, who wanted to say that Network File Systems (NFS) has moved from a checkbox feature to a design decision that shapes the entire storage offering. He reckons that, for decades, NFS has been treated as “plumbing”, something every storage solution simply “has.” However, for storage vendors and system integrators, how you deliver NFS is now a strategic choice that directly impacts performance, cost, and your ability to compete. He sees growing demand for unified NFS + SMB access in enterprise environments, and Tuxera products such software. We asked Ned a set of questions to find out more. Blocks and Files: Tell me about Tuxera and its main products Ned Pyler: Tuxera has spent decades building storage and networking software for embedded and enterprise systems. In the enterprise space, Fusion SMB represents the cutting edge of SMB 3 support for Linux, with the best performance, scale, and reliability in the industry. Fusion SMB has provided mission-critical remote file system support to IBM, Quantum, and other storage providers for nearly 10 years. Now, with Fusion NFS, the ideal Linux-based server for the 4.x protocol is coming to market this year. Blocks and Files: How has the role of NFS evolved? Please identify modern and legacy approaches. Ned Pyler: NFS is the most common standards-based remote file server for Unix-like systems; many other proprietary remote storage systems have come and will go, but NFS will remain. It evolved from humble capabilities in the 1980s to now, using networks like RDMA and security mechanisms like Kerberos. In the modern enterprise era, Linux has become king, but files never stopped being the most common organisational data, and NFS is the common glue for remote access to files between Linux systems. Blocks and Files: How has the role of SMB evolved? Please identify modern and legacy approaches. Ned Pyler: SMB began at the same space as NFS, a simple workgroup remote file system. In its early days, SMB was the platform of Windows, given little attention by Unix systems admins. It later became the primary system of macOS and, in time, became heavily integrated into the Linux kernel. While not standards-based, it became open, and many third parties adopted its use. Microsoft began real innovation in SMB 3, adding extreme performance, security, access, and reliability capabilities like multichannel, RDMA, QUIC, encryption, persistent handles, compression, and scale-out clustering. SMB 3.1.1 is as different from SMB’s beginnings as a Formula 1 car is from a horse cart. Blocks and Files: What influence on file services has come from containerisation and software-defined storage? Ned Pyler: Containerization signalled the end of “kernel-is-supreme” remote file servers. To get good performance 20 years ago, NFS and SMB servers needed to run in kernel, and that’s what Windows and Linux NFSD do. But containers need user-mode software to run safely and securely. That means Windows Server cannot be a container platform for SMB, nor can NFSD for NFS. Modern features are what power software-defined storage systems, allowing them to abstract hardware and end the stranglehold of vendor appliance solutions and proprietary hardware. SDS is what powers modern private cloud computing, hosting, AI-ML training, and the 21st-century enterprise; SMB 3 itself was designed primarily as a fabric protocol for SDS and is often used in Microsoft technologies, not a file system, but as a data block transport for technologies like Storage Spaces Direct, Cluster Shared Volumes, Storage Replica, and Hyper-V Live Migration. Linux open-source options like Ganesha and Samba don’t offer modern features like RDMA and scale-out, making them less viable in SDS. Blocks and Files: Could you explain the differences between proprietary, kernel-based, and user-space NFS architectures? What are the trade-offs between them? Ned Pyler: Kernel-based file servers live as modules or drivers within a shared memory space along with other software and the operating system itself. In older times and older processors, kernel was the only way to squeeze maximum performance. Its use now is legacy. Kernel-based software is inherently more dangerous, where any crash will interrupt the entire server, and any security compromise allows complete takeover. User-mode operates in the safer world of process isolation, where a file server crash only stops the file server. In many cases, it meant comparative slowness, but between the improvements in processors, hardware, coding techniques, and operating system optimisations, the right user-mode file server can perform just as well or even better than a comparative kernel-mode one. Blocks and Files: Why do you think legacy NFS approaches are no longer fit for modern workloads? So what needs to be done? Ned Pyler: The common open source NFS servers are constrained by their designs and nature. NFSD has decent performance, but living in kernel means it’s too dangerous and too inflexible. Ganesha is much safer and can be securely containerized, but lacks performance and doesn’t allow the use of modern technologies. File servers don’t need to compromise anymore; they can be fast, scalable, highly available, reliable, and secure, all while running in user-mode, taking advantage of common technology like RDMA networking and Direct IO flash, and rapidly innovating as new technologies become available. The NFS 4.x protocol itself provides the capabilities modern workloads need, but without the right features and engineering teams behind it, it isn’t going to shine. Furthermore, SMB 3 has advantages that traditional Linux NFS operators may have overlooked; for instance, if you run the MLPerf 3D-Unet benchmark, the most stressful AI training and checkpointing option for storage, you’ll find that a file server and client running SMB 3 will perform significantly better than when running NFS 4. That SMB protocol was built as a high-performance data mover fabric; it’s just better at the kind of IO that AI training creates. Blocks and Files: Is there demand for unified NFS + SMB access in enterprise environments? Why is this? Ned Pyler: While Linux and macOS both have solid modern NFS and SMB clients, Microsoft stopped developing theirs in Windows 2000. Furthermore, Windows and Mac use SMB as a core protocol for many operations, especially with entrenched Active Directory tie-ins like Group Policy, logon scripts, and home folders. The SMB client arrived as a first-class solution to Linux very late, and regardless, the mindshare was already firmly with NFS. This means that mature organisations have two choices when it comes to file data. They can either completely replace the use of one with the other, which would likely mean a complex and expensive ending to all use of NFS if they were a mixed Windows-Unix org, or simply use a file server that supports safe client access to both. Because each protocol has very different security and data locking semantics, you can’t simply share the same data with two file servers; it requires true multiprotocol understanding, and systems like Tuxera Fusion provide this safe, inexpensive, and easily managed offering. Blocks and Files: What is the future of file services? Ned Pyler: The most important thing to understand is that files are not going anywhere. Despite decades of databases, REST, blobs, and other formats in the enterprise, the most likely place for data to live is in a file. The files keep getting bigger, and more are created than deleted. Most of what you’d call the “product” of enterprise organisations is simply files. As technology continues to evolve and hopefully improve ways we cure disease, grow food, manage logistics, or reduce human errors, files will continue to be a primary way to preserve and share knowledge. The problems and solutions will continue to be around speed, size, and reliability as the endless flow of files continues to pour out of people and their tools. Files and file servers will outlive us all.
document fetched 2026-06-06 08:27