MaxLinear and Los Alamos National Laboratory Jointly Advance High-Performance File System Acceleration for HPC Storage
MaxLinear, Inc. (NASDAQ: MXL), a leading provider of high-performance storage accelerator SoCs, and Los Alamos National Laboratory (LANL) have announced a collaboration to enable hardware-accelerated OpenZFS File System storage for large-scale, high-performance computing (HPC) environments. The two organizations have jointly developed an architecture designed to dramatically improve throughput and storage capacity for next-generation NVMe flash-based storage infrastructure.
At the heart of the collaboration is MaxLinear's Panther™ Storage Accelerator, integrated with ZFS as a Data Processing Unit Services Module (DPUSM) provider. This enables inline hardware acceleration of CPU-intensive operations — including data compression and checksum generation — to increase storage capacity, improve file I/O performance, and reduce host CPU utilization while fully preserving ZFS ordering, consistency, and data integrity guarantees.
The results are striking: without Panther™, ZFS is limited to approximately 1.2 GB/s writes and 8.1 GB/s reads. With hardware offload, the collaboration achieved 57 GB/s reads and 47 GB/s writes with GZIP L9 compression at a roughly 1.3:1 ratio — representative of high-entropy scientific data — delivering write speedups of approximately 39x and read speedups of approximately 7x.
LANL's Role: Direct I/O and the ZFS Interface for Accelerators
LANL brings decades of experience operating ZFS at scale for national security, science, and energy computing workloads. Over that time, the laboratory's teams have led development of critical filesystem extensions — most notably Direct I/O support and the ZFS Interface for Accelerators (ZIA), a structured framework for introducing hardware acceleration into the ZFS data path without modifying core filesystem semantics.
ZIA is the enabling architecture that makes this collaboration possible: it allows multiple Panther™ Storage Accelerators to be deployed in parallel, scaling performance without introducing serialization or centralized bottlenecks. It also ensures that the data integrity guarantees that ZFS is renowned for — ordering, consistency, checksumming — remain fully intact regardless of the offload path taken.
"Los Alamos' Direct I/O support and Z.I.A. (ZFS Interface for Accelerators) work were developed to accelerate performance for the ZFS-using community. In this collaboration, MaxLinear demonstrated hardware-offloaded ZFS operations with reported speedups of approximately 39x for writes and 7x for reads. These results illustrate the potential for accelerator-based approaches to reduce host CPU involvement while maintaining the data-protection benefits associated with ZFS."
— Gary Grider, Senior Director for Computing Technologies, LANLMaxLinear's Contribution: The Panther™ Storage Accelerator SoC
MaxLinear contributes the Panther™ family of Storage Accelerator SoCs alongside Storage Software Development Kits, providing high-throughput, low-latency execution of ZFS data path services using a domain-specific high-performance SoC architecture. Panther™ executes deep data compression, encryption, deduplication, and data protection services entirely inline in hardware — significantly reducing host CPU overhead while maintaining the full pipeline performance expected by HPC workloads.
The Panther™ integration supports multi-hundred gigabit scalability, and its modular deployment model means organizations can add accelerators in parallel to scale throughput further — without architectural rework or operational disruption to existing ZFS-based storage environments.
"Los Alamos National Laboratory has been at the forefront of advancing storage architectures for high-performance computing. By enabling hardware-accelerated ZFS with Panther™ Storage Accelerators, we deliver deep data compression, data protection services, and multi-hundred gigabit scalability—while preserving the data integrity guarantees that ZFS is known for."
— Vikas Choudhary, EVP of Connectivity & Storage, MaxLinearKey Capabilities of the Hardware-Accelerated OpenZFS Architecture
The joint collaboration delivers three primary capabilities that address the core bottlenecks in traditional ZFS deployments for large-scale scientific and enterprise computing environments.
Hardware-Assisted Deep Data Compression
Compression offload removes the dominant CPU burden on high-throughput I/O paths. Panther™ executes GZIP-level compression entirely in hardware, enabling high I/O performance with minimal impact on host CPU utilization — critical for HPC environments where CPU cycles are precious compute resources shared across many workloads simultaneously.
Scalable Parallel Accelerator Deployment
Multiple Panther™ Storage Accelerators can be deployed in parallel through the ZIA framework, enabling organizations to scale storage performance linearly without introducing serialization bottlenecks or centralized throughput chokepoints. The architecture is designed to grow with the infrastructure rather than require wholesale replacement as data volumes increase.
High-Bandwidth Operation at Scientific Data Scale
Demonstrated for the first time, the architecture achieves 57 GB/s reads and 47 GB/s writes with GZIP L9 compression — conditions representative of high-entropy scientific datasets. Standard ZFS without Panther™ is limited to approximately 1.2 GB/s writes and 8.1 GB/s reads under the same workload, making the ~39x write and ~7x read improvements transformative for petabyte-scale HPC storage deployments.
The significance of this announcement extends beyond the benchmark numbers. HPC storage environments have long faced a fundamental tension: ZFS offers unmatched data integrity and filesystem maturity, but its CPU-intensive operations — compression, checksumming, deduplication — have historically made it difficult to scale to the multi-hundred gigabit throughput that modern NVMe flash arrays can theoretically sustain. This collaboration directly addresses that gap, enabling organizations to run ZFS at its full potential without sacrificing compute resources that scientific workloads need.
As AI and scientific computing workloads continue to drive explosive growth in storage requirements, hardware-accelerated approaches like this collaboration represent a credible path to storage infrastructure that can keep pace — scaling performance, preserving data integrity, and freeing CPU resources for the computations that matter most.
For complete technical details on the Panther™ Storage Accelerator family and the OpenZFS hardware acceleration architecture, visit MaxLinear's official Panther™ product page.