/
2018-03-20 HKG18 HPC Meeting Minutes

2018-03-20 HKG18 HPC Meeting Minutes

Nov 07, 2018
[ 1 Attendance ] [ 2 Agenda  ]

Attendance

Committee Members 

Name

Present

Name

Present

Kanta Vekaria (HPC Lead, OCTO Linaro)

Martin Stadtler (Director of LEG, Linaro)

Darren Cepulis (ARM)  

 

Andrew Wafaa (ARM)

Tianhong Ding /Joshua Mora (Huawei)

 

Larry Wikelius (Cavium)

Scott Hara (Qualcomm) 

Steve Geist (Qualcomm)

 

Hongbo Zhang (HXT)

Xin-Xin Yang (HXT)

Jon Masters (RedHat)

Takeharu Kato (Fujitsu)

Koichi Hirai (Fujitsu)

 

Guests 

Name

Present

Name

Present

David Rusling (CTO, Linaro) 

Andrea Gallo (VP of Segment Groups, Linaro)

 

Anoop Saxena (Project Manager, Segments)

Renato Golin, Don Harbin (Linaro)

 

Shen (Jackie) Hui, Kang-Kang (Huawei)

 

Richard Henwood (Arm)

Yan Fisher (RedHat)

 

Agenda 

 

  • Quick Reminder

    • QEMU patches upstream, please test, verify, ACK

  • Filesystems: GPFS, Lustre

  • Growing Engineering in the SIG

    • Incentivising participation and collaboration

  • AOB

 

Minutes

Filesystems

  • Larry: Filesystems in Arm are not currently discussed. GPFS is an IBM closed offering. Lustre is a leading HPC filesystem. Would like Renato to ensure that the OpenHPC Lustre client is validated on Arm and at parity with other architectures.

  • Lustre

    • Has a strong influence by Intel. 

    • We need to understand its scoping for Arm and where we can influence

    • Oakridge working on CI/CD upstream of Lustre

    Lustre user Group April 23rd Argon. Who is participating?

  • Lustre Server

    • Likely a bigger issue

    • There are risks around the underlying filesystem (ZFS, BeeGFS uses ZFS)

  • Lustre Client

    • Start here on OpenHPC

    • OpenHPC has a Luster server in TACC, which currently runs for Intel and not Arm

  • Other notes

    • Ceph is replacing NFS

    • Fujitsu: Testing with NFS

    • RedHat: Don't support Lustre and have zero interest. Lustre == Oakridge. We should focus on what the HPC labs find as important 

    • CephFS works just as well as Lustre

    • Cavium: 

      1. Not asking for distros to pick this up

      2. Lustre is a well established setup that needs supporting

      3. We need to agree what that 'something' is which we support for Arm

    1. Collate information on what labs are using. Need to collate as a sig

    2. If Lustre is the way to go then run Lustre Client on OpenHPC to identify scale of issues

 

Growing Engineering in the SIG

  • Compiler optimisation lead project needs staffing

  • Reference Implementation discussion

    • RedHat compatible RPM for openHPC

  • Continue Joint marketing and business development. Key targets this year

    • ISC18

    • SC18

  • Key messages

    • Engineering collaboration

    • Business Development, e.g. getting labs and others to join the SIG and collaborate 

    • Marketing

  • Thank you to Fujitsu, Cavium, Arm and RedHat for putting themselves forward. I (Kanta) will be working closely with you to make this happen firstly for ISC18 in Frankfurt