Certification notes for the SGI Altix UV 100 and Altix UV 1000 on Red Hat Enterprise Linux 6

Updated -

Red Hat Enterprise Linux 6.0

The SGI Altix UV 100 and Altix UV 1000 are certified and fully supported on Red Hat Enterprise 6.0 with up to 128 logical processors and 2TB of memory (physical configuration).

Both systems also require the following kernel command line parameters with Red Hat Enterprise Linux 6.0:

virtefi nobau nmi_watchdog=0

Red Hat Enterprise Linux 6.1

The SGI Altix UV 1000 has been certified and is fully supported on Red Hat Enterprise Linux 6.1 with up to 1280 logical CPUs when using the Intel® Xeon® E7 processor family and up to 8TB of memory. The following kernel command line parameters were used when certifying this configuration:

nmi_watchdog=0 add_efi_memmap nortsched processor.max_cstate=1 log_buf_len=8M pci=hpiosize=0,hpmemsize=0,nobar nohz=off cgroup_disable=memory pcie_aspm=on intel_iommu=off

These parameters will typically be configured automatically by the SGI Foundation Software bundle provided by SGI. Specific details are as follows:

  • nmi_watchdog=0 - Disables the NMI watchdog to maximize available CPU time on large SSI configurations.
  • add_efi_memmap - Allows for extra memmap entries, which is required for large SSI configurations..
  • nortsched - Prevents livelocks during boot on large SSI configurations. 
  • log_buf_len=8M - Increase the size from the default 128k, which is required for large SSI configurations.
  • processor.max_cstate=1 - Optimal cstate setting for large configurations.
  • pci=hpiosize=0 - Required to support PCI hotplug on systems with multiple Magma chassis.
  • hpmemsize=0 - Required to support PCI hotplug on systems with multiple Magma chassis.
  • nobar - Disable assigning a resource to a device BAR if that BAR was not assigned by the BIOS.

  • nohz=off  - Avoids excessive load balancing on large systems

  • cgroup_disable=memory - Disable memory resource controller to maximize available memory.

  • pcie_aspm=on - Enables ASPM, which previously had to be disabled to workaround a bug.

  • intel_iommu=off - Disables the IOMMU due to DMA address mask bugs with certain hardware

Comments