2017-06-08 LEG SC Meeting Agenda/Minutes

Owned by Former user (Deleted)
Last updated: Jan 08, 2018 by Don Harbin


Attendance



 


   

Live meeting notes will be  https://docs.google.com/document/d/1Ab3Vh-tCDQDLlXKhNH-4xxfx0S8YoqDCNqr4mVEbMfM/edit   

Agenda

  • Health Checks

  • SBBR / SBSA

  • Member Meeting July (Tokyo)

  • AOB

Minutes

'''Present'''
Martin
Masahisa Kojima
Satoru Okamoto
Jon Masters
Takeharu Kato
Koichi Hirai
Graeme
Kanta
Anoop


=Minutes: Actions: Statements: Decision=

@action
@statement
@decision

Martin @statement: Health Card : Monthly sent to all and in parallel propose to review with members between the connect

Jon @statement : BigData, which CPU bug? Which Silicon?
Martin @action: Will provide more details

Jon ; Very keen on SDEI and RAS in 4.13 but I am not certain they will get in With the IORT rev C you will be able to boot Huawei and Cavium out of the box with SMMUv3 support

Martin @statement: 4.12, pull request window will be close by tomorrow.

Jon @statement: vsmmu support is needed for DPDK within containers under VMs.
That's the last missing core piece for OpenStack hardware enablement. Many other VM pieces are missing but not critical, including NUMA support within guests, and limits for guests - not CPU/memory hotplug and a limit of 112 vcpus

Message from Jon Masters: I am personally very keen on SDEI and RAS in 4.13 but I am not certain they will get inI am personally very keen on SDEI and RAS in 4.13 but I am not certain they will get in
Message from Jon Masters: With the IORT rev C you will be able to boot Huawei and Cavium out of the box with SMMUv3 support :)With the IORT rev C you will be able to boot Huawei and Cavium out of the box with SMMUv3 support
Message from Jon Masters: vsmmu support is needed for DPDK within containers under VMsvsmmu support is needed for DPDK within containers under VMs
Message from Jon Masters: That's the last missing core piece for OpenStack hardware enablement. Many other VM pieces are missing but not critical, including NUMA support within guests, and limits for guests - not CPU/memory hotplug and a limit of 112 vcpusThat's the last missing core piece for OpenStack hardware enablement. Many other VM pieces are missing but not critical, including NUMA support within guests, and limits for guests - not CPU/memory hotplug and a limit of 112 vcpus

Anoop: to create collaborate page for above points “ARM64/X86 Parity Feature Tracking”

Jon @Action ask all to identify any blockers

Kanta @statement SVE spec published.

Jon @statement: there are 2# HPC conferences
Supercomputer in US
International supercomputer in Frankfurt in Nov’’17

Follow on email from JonM

As I noted on the call, there are a couple of missing virt pieces needed. Broadly these fall into minimum enablement and x86 party boxes.

The former is mostly:

1. Enable vsmmu for guests to run DPDK hosted instances of containers.

The latter is mostly:

1. Enable NUMA support in guests
2. Enable CPU and memory hotplug
3. Enable RAS events for guests
4. Rework vGIC to support more than 112 vCPUs

There are a few more. The LEG leads are going to create a Collaborate page that we can throw things onto to get to x86 parity soon.

The vsmmu piece is pretty important. Eric Auger from RH is working on this alongside ARM but it will be needed for a full NFV solution. Currently, we are able to do hardware passthrough into guests, but not into containers within guests. So we can enable many use cases, but not all. At home I have a 10G Intel NIC running inside a VM and will be poking at DPDK next - some of the RH team have seen really bad performance numbers and this needs further triaging with "real world" PCI hardware of the kind end users deploy, such as the higher end Intel NICs. I also look forward to playing with QLogic and Mellanox adapters more.


  • @ name action item here //dd-mm-yyyy  

<end>