User:FlyingBlackbird

From RCS Wiki
Revision as of 03:00, 9 February 2020 by FlyingBlackbird (talk | contribs) (Add R)
Jump to navigation Jump to search

Intended Usage

I am planning to use a Blackbird Desktop System with an IBM Power9 v2 (= stepping DD2.3) 8-core CPU as testing infrastructure for open source projects (mainly https://www.r-project.org/ and related packages).

Instable hardware watchpoints in stepping DD2.2 (erratum #1: DAWR [Data Access Watchpoint Register] feature is disabled on DD2.2) was the main reason for me to wait for DD2.3 to have reliable low-level debugging available.

System configuration

Component Brand Model Costs in EUR
Mainboard Raptor Blackbird Rev. 1.01
CPU IBM 8-Core POWER9v2 with 3U HSF
Desktop Case Fractal Design Define R6 USB-C 140
Power Supply be quiet! Straight Power 11 (650 W) 120
RAM Samsung (OEM by phs memory) 2 x M393A4K40CB2-CTD7Q 32 GB DDR 4 RDIMM 344
Video Card Aspeed AST 2500 onboard VGA 0
Optical Drive Asus BW-16D1HT Retail (Blu Ray Writer) 75
HDD Seagate IronWolf Pro 8 TB (ST8000NE0004) SATA III 300
Operating System Ubuntu Server 19.10 0
Operating System Fedora Server 31 0
Total

Measures

All measures were observed with a minimal system setup (only HDD with Ubuntu Server 19.10 and the built-in AST2500 GPU).

Power Consumption:

  • 2.2 W plugged in
  • 68 W Idle
  • 186 W with 100 % CPU load (testet with stress --cpu 32 -t 180s)

Temperatures:

  • CPU idle: 45 degree Celsius
  • CPU 100 % load: 72 degree Celsius
  • DIMM 0 (32 GB ECC RAM): 44 Degree Celsius
  • DIMM 1 (32 GB ECC RAM): 49 degree Celsius

Noise Level:

  • Surprisingly quiet during idle and normal use (low CPU usage)
  • Fans spinning fast but not too loud during 100 % CPU load (slow RPM increase and decrease, no annoying sudden changes)

Boot-up durations

  • Cold boot (powered off)
    • Switching on the power supply switch: Time until the case power switch reacts to start Hostboot: 2 minutes
    • After cold boot: Time until Petitboot boot menu appears: 2 minutes
    • Booting Ubuntu 19.10 Server on a NVMe SSD: Time from Petitboot boot menu to Ubuntu lightdm login screen: 34 s
  • Soft boot (powered on but switched off via the case power switch)
    • Time until the Petitboot boot menu appears: about 100 seconds
  • Reboot from Ubuntu until Petitboot boot menu appears: about 4 - 10 seconds (after Ubuntu has shut down)

Summary:

  • A cold start takes quite long compared x86_64 architecture (about 4,5 minutes until the Ubuntu login screen appears)
  • A warm start does also take longer compared x86_64 architecture (about 2 minutes until the Ubuntu login screen appears)


Software Installations

Ubuntu Server 19.10

  • Install Gnome (or another Desktop) via tasksel:
    • sudo apt-get update && sudo apt-get upgrade
    • sudo apt-get install tasksel
    • Start tasksel in a shell
    • choose "Ubuntu desktop" to install Gnome.
  • If Gnome does hang after booting (most probable): Install lightdm as display manager (replacing gdm3):
    • Reboot into recovery via petitboot and open the root shell
    • sudo apt-get install lightdm
    • Reboot

Test status

Hardware being tested at the moment (but not yet fully successfully in the complete system due to NVMe problems when installing Linux)

  • NVMe M.2 SSD together with a SATA HDD and Ubuntu 19.10 Server
    • RaidSonic ICY BOX IB-PCI214M2-HSL M.2 to PCIe adapter
    • Samsung 970 EVO Plus (M.2 with NVMe) 2 TB

Findings & Solutions (work in progress)

SQASHFS error logged during Ubuntu Server 19.10 Installation

These error messages are shown during the installation process and seem to be independent of the actual hardware configuration:

SQASHFS error: xz decompression failed, data probably corrupt
SQASHFS error: squashfs_read_data failed to read block 0x0
SQASHFS error: Unable to read metadata cache entry [0]

NVMe SSD is recognized but the installation seems to use the wrong device name later:

Error: Could not stat device /dev/disk/by-id/wwn-eui.0025... - No such file or directory
...
An error occured handling 'disk-nvme0n1': OSError - [Errno Failed to find device at path: %s] /dev/disk/by-id/wwn-eui.0025...
...
Traceback:
...
File "/snap/subiquity/1286/lib/python3.6/site-packages/curtin/commands/block_meta.py", line 182, in devsync

Workaround to install Ubuntu 19.10 Server on an NVMe SSD: Follow the installer until you have specified the partitions but do not yet hit Done to apply this workaround:

  • Press Ctrl+Alt+F2 to enter a console outside of the installation program.
  • If you are not already in a root shell: Login with the user created earlier in the installation process
  • Find out your SSD ID
/dev/disk/by-id/
  • Create a symbolic link to your SSD (try with and without "nvme-" and prefix, eg. "nvme-eui.0025... and "eui.0025...":
# If you have more than one NVMe SSD you may have to change device number eg. to <code>/dev/nvme0n2</code> for the second SSD...
sudo ln -s /dev/nvme0n1 /dev/disk/by-id/wwn-eui.0025...
  • Exit the shell by pressing Ctrl+D or typing exit.
  • Switch back to the installer using Ctrl+Alt+F1.
  • Now choose "Done" to apply and use the partitions. The installer will continue the installation on this NVME now

Failed to install grub (cannot find GRUB drive for /dev/nvme0n11)

Error message:

Failed to install grub!
grub-install: error: cannot find GRUB drive for /dev/nvme0n11.  Check your device.map.

This may probably only happen if you more than one PReP partition on your system (eg. one on the SSD and one on the HDD).

Currently this error can be ignored but the installer does not finish its work. So you have to

  • Boot into a root shell via the recovery boot entry in Petitboot
  • Give your user sudo rights (sudo adduser <login_name> sudo)
  • Then reboot and continue with the installation of the GUI (eg. Gnome)