WOA Issue 70

on Posted on
Works on Arm News, October 26, 2018.

In this issue

  • Arm Neoverse roadmap for server-class systems
  • Linux kernel to support big/little process scheduling (Phoronix)
  • Fedora 29 on its way
  • NetBSD ARM64 images for SBSA compliant systems
  • ARMv8.3 has an instruction just for Javascript

Arm Neoverse roadmap for server-class systems

At Arm TechCon, Arm announced a roadmap for server-class systems called Neoverse. The company plans to release IP that will enable licensees to build fast multi-core systems without going through the expense and bother of designing their own CPU cores. The roadmap includes 7nm and 5nm parts with an annual cadence for improvements.

Neoverse is part of a broader strategy that identifies opportunities for Arm based systems, with a “cloud to edge infrastructure foundation for a world of one trillion intelligent devices”.

Linux kernel to support big/little process scheduling (Phoronix)

The plan is to better handle “misfit” tasks, which need more CPU but are currently scheduled on a slower “little” core. The patch lands in the 4.20 kernel. This is primarily targeted at CPU cores that incorporate fast high speed cores with slower but more memory efficient secondary cores.

Fedora 29 is on its way

Fedora 29 is due to be shipped live on Tuesday, October 30, 2018. Improvements in the release target IOT support for smaller 32-bit and 64-bit Arm devices, including newly improved ZRAM support for swap in memory constrained devices. Fedora 29 is one of the first operating systems also to support Go 1.11, which has substantial performance improvements over earlier Go releases on arm64 especially on crypto codes.

NetBSD ARM64 images for SBSA compliant systems

Jared McNeill has been working on code to boot NetBSD cleanly on SBSA compliant systems. SBSA (“Server Base System Architecture”) is a standard for arm64 server designs, with the goal that a server will be able to support booting a wide range of operating systems without the OS vendor needing to customize the boot code for each vendor hardware.

ARMv8.3 has an instruction just for Javascript

The FJCVTZS (“Floating-point Javascript convert to signed fixed-point, rounding toward zero”) instruction is new in the ARMv8.3 instruction set. The Javascript language lacks an integer type, so this routine efficiently handles a common float-to-int with appropriate semantics.

To take advantage of this, you’ll need support in your CPU (like the latest Apple A12) as well as in your Javascript interpreter or JIT. Where it’s implemented, codes show an overall 0.5% to 2% improvement, with crypto codes like AES and SHA256 showing the biggest speedups.

RT : CXL is important for our industry and is thrilled to help drive it. community.arm.com/developer/ip…

RT : The cases arrived from factory today; perfect fit, excellent finish and a smooth aluminum alloy injectio… https://t.co/urmT5TOrJe

RT : With this small change my k8s controller startup went from seconds to milliseconds :) github.com/kubernetes/kubernet…