Phoronix reports that the upcoming Linux 6.7 kernel will introduce a new feature for Intel’s QuickAssist Technology (QAT) device driver. This feature will enable more efficient use of QAT Gen4 hardware, specifically the latest-generation Intel Xeon Scalable “Sapphire Rapids” processors.

The QAT Gen4 devices, which are currently available with select Sapphire Rapids processors, support chained compression “DC” operations. With this feature, a single firmware request can perform both hashing and compressing of data in one step.

The Linux kernel driver for Intel QAT is being extended to support the DC chaining service with the latest QAT accelerators. By allowing the data to be hashed and compressed in a single request, this feature reduces overhead and latency compared to making multiple requests to the QAT hardware.

Intel QAT is a hardware-based technology developed by Intel to accelerate cryptographic and data compression workloads. It offloads and accelerates specific operations, such as encryption, decryption, and compression, from the CPU to dedicated hardware accelerators. This improves performance and efficiency while freeing up CPU resources for other tasks.

Key features and benefits of Intel QAT include cryptographic acceleration, data compression, improved performance, reduced CPU utilization, hardware security, virtualization support, and various form factors such as PCIe cards and integrated solutions.

This new feature in Linux 6.7 will be of interest to those using home servers, as it allows for more efficient use of Intel QAT Gen4 hardware. By offloading cryptographic and compression tasks to dedicated hardware accelerators, home servers can experience significant performance improvements and reduced CPU utilization. This can lead to better handling of transactions and connections, lower power consumption, and improved overall system efficiency.

Source: Phoronix.

Linux 6.5 now defaults to the AMD P-State EPP driver for Zen 2 and newer Ryzen systems, as long as the system supports ACPI CPPC. However, the AMD EPYC server processors still continue to use ACPI CPUFreq by default. Given the increasing interest in the AMD Ryzen 7000 series for budget and small-to-medium-sized business (SMB) servers, the performance impact of Linux 6.5 with more server workloads was analyzed.

Phoronix has tested the changes, and testing was conducted comparing the performance of Linux 6.4 against Linux 6.5, both out-of-the-box and using the Ubuntu Mainline Kernel PPA for easy reproducibility. The default change involves going from ACPI CPUFreq Schedutil to AMD P-State EPP with the powersave governor. Additional tests were done with the performance governor for maximum performance. AMD P-State is available on earlier Linux kernel versions but is not set to be used out-of-the-box until Linux 6.5 and later. The testing was done using the ASRock Rack 1U4LW-B650/2L2T, a 1U Ryzen AM5 server platform that supports Ryzen 7000 series processors and ECC memory. No other changes were made to the server during testing, except for swapping out the Linux kernel and running secondary tests with the performance governor. The CPU clock frequency differences in the automated system table were minimal and did not affect the testing results.

The article provides valuable insights for those interested in using Ryzen processors for server applications.

Source: Phoronix.

PuzzleFS has been quietly making progress as a new file-system designed specifically for containers, writes Phoronix. Developed by Cisco engineers, PuzzleFS aims to address the limitations of the OCI (Open Container Initiative) and is written in the Rust programming language.

The kernel driver for PuzzleFS, also written in Rust, is currently being developed outside the mainline Linux kernel. This is due to the lack of necessary Rust abstractions in the mainline kernel. PuzzleFS offers several design goals, including immutability, reduced duplication, reproducible image builds, direct mounting support, data integrity, and memory safety guarantees. The file-system also includes optimal Zstd compression support.

Source: Phoronix.

NVIDIA is making advancements in their Linux InfiniBand driver to support 800Gb/s (XDR) speeds. The company has posted a set of six patches for the upstream Linux kernel, aiming for inclusion in Linux 6.7. These patches will enable the 800Gb/s support in the InfiniBand core networking code and the Mellanox MLX5 driver. The NVIDIA Mellanox Skyway GA100 InfiniBand to Ethernet Gateway is believed to be the first product capable of achieving the XDR speeds. This development showcases NVIDIA’s commitment to delivering high-speed networking solutions for servers.

The details of the preparations for the 800Gb/s (XDR) speed implementation in the Linux kernel’s InfiniBand code can be found in the patch series provided by NVIDIA.

Source: Phoronix.

The Linux 6.6 kernel version seems to be delivering significant performance improvements for AMD EPYC server CPUs according to Phoronix. Tests conducted on Genoa and Genoa-X processors as well as Intel Xeon Scalable “Sapphire Rapids” processors showed that the performance of AMD EPYC servers was greatly enhanced with Linux 6.6 compared to Linux 6.5 stable. The improvements were especially notable for certain workloads.

The Linux 6.6 kernel introduces the EEVDF scheduler and workqueue enhancements that benefit chiplet-based CPUs with multiple L3 caches, such as those used by AMD. Overall, the new features in Linux 6.6 seem to be positively impacting AMD server CPU performance.

Source: Phoronix.

Phoronix is reporting that Linux 6.6-rc2 has been released by Linus Torvalds, coinciding with the 32nd anniversary of the introduction of the Linux 0.01 kernel version. Torvalds notes that this release does not have any standout features, but includes various fixes and changes in drivers, along with some tracing fixes and other minor updates.

Source: Phoronix.