OS Tuning
submit= MYMASK=`printf '0x%x' \$((1<<\$SPECCOPYNUM))`; /usr/bin/taskset \$MYMASK $command
When running multiple copies of benchmarks, the SPEC config file feature submit is sometimes used to cause individual jobs to be bound to specific processors. This specific submit command is used for Linux. The description of the elements of the command are:
Using numactl to bind processes and memory to cores
For multi-copy runs or single copy runs on systems with multiple sockets, it is advantageous to bind a process to a particular core. Otherwise, the OS may arbitrarily move your process from one core to another. This can effect performance. To help, SPEC allows the use of a "submit" command where users can specify a utility to use to bind processes. We have found the utility 'numactl' to be the best choice.
numactl runs processes with a specific NUMA scheduling or memory placement policy. The policy is set for a command and inherited by all of its children. The numactl flag "--physcpubind" specifies which core(s) to bind the process. "-l" instructs numactl to keep a process memory on the local node while "-m" specifies which node(s) to place a process memory. For full details on using numactl, please refer to your Linux documentation, 'man numactl'
numactl --interleave=all "runspec command"
Launching a process with numactl --interleave=all sets the memory interleave policy so that memory will be allocated using round robin on nodes. When memory cannot be allocated on the current interleave target fall back to other nodes.
Transparent Huge Pages
Transparent Hugepages increase the memory page size from 4 kilobytes to 2 megabytes. Transparent Hugepages provide significant performance advantages on systems with highly contended resources and large memory workloads. If memory utilization is too high or memory is badly fragmented which prevents hugepages being allocated, the kernel will assign smaller 4k pages instead.
On RedHat EL6 and later, Transparent Hugepages are used by default if /sys/kernel/mm/redhat_transparent_hugepage/enabled is set to always. The default value is always.
On SUSE SLES11 and later, Transparent Hugepages are used by default if /sys/kernel/mm/transparent_hugepage/enabled is set to always. The default value is always.
ulimit -s [n | unlimited] (Linux)
Sets the stack size to n kbytes, or unlimited to allow the stack size to grow without limit.
KMP_STACKSIZE=integer[B|K|M|G|T] (Linux)
Sets the number of bytes to allocate for each parallel thread to use as its private stack. Use the optional suffix B, K, M, G, or T, to specify bytes, kilobytes, megabytes, gigabytes, or terabytes. The default setting is 2M on IA32 and 4M on IA64.
KMP_AFFINITY=physical,n (Linux)
Assigns threads to consecutive physical processors (for example, cores), beginning at processor n. Specifies the static mapping of user threads to physical cores, beginning at processor n. For example, if a system is configured with 8 cores, and OMP_NUM_THREADS=8 and KMP_AFFINITY=physical,2 are set, then thread 0 will mapped to core 2, thread 1 will be mapped to core 3, and so on in a round-robin fashion.
OMP_NUM_THREADS=n
This Environment Variable sets the maximum number of threads to use for OpenMP*
parallel regions to n if no other value is specified in the application. This
environment variable applies to both -openmp and -parallel (Linux)
or /Qopenmp and /Qparallel (Windows). Example syntax on a Linux system with 8
cores:
export OMP_NUM_THREADS=8
Default is the number of cores visible to the OS.
vm.max_map_count=-n (Linux)
The maximum number of memory map areas a process may have. Memory map areas are used as a side-effect of calling malloc, directly by mmap and mprotect, and also when loading shared libraries.
intel_idle.max_cstate=n kernel boot option (Linux)
Linux has a kernel boot option to control C-States - similar to how C-States can be controlled through BIOS options. This feature selects the processor's lowest idle core power state (C-state) which the operating system will utilize. The higher the C-State, the lower the power usage of that idle state (3 is the lowest power idle core state supported by the processor). Values for this setting can be:
PM Quality of Service (PM QoS) Interface (Linux)
This interface allows for more fine-graned control of power saving states through the kernel and a user mode interface. Both instances of this tool allow for setting performance expectations of drivers, subsystems, and applications through one of the parameters. The parameters available to the PM QoS include classes for cpu_dma_latency, network_latency, network_throughput, memory_bandwidth as well the ability to manage per-device latently contraints and PM QoS flags.
The cpu_dma_latency allows a user to set specific target latencies for the CPU, allowing for indirect control of the C-state residency of processors in a system. Setting this to 0 will prevent transitions to deep sleep states. Setting it to other values will allow a system to enter various c-states, depending on the architecture.
Free the file system page cache (Linux)
The command "echo 1 > /proc/sys/vm/drop_caches" is used to free up the filesystem page cache.
Performance Governors (Linux)
The in-kernel CPU frequency governors are pre-configured power schemes for the CPU. The CPUfreq governors use P-states to change frequencies and lower power consumption. The dynamic governors can switch between CPU frequencies, based on CPU utilization to allow for power savings while not sacrificing performance.
For the Performance Governor the CPU frequency is statically set to the highest possible for maximum performance.
On SUSE SLES 11 and 12 systems you can set the in-kernel CPU frequency governor for all CPUs to the Performance Governor with the following command:
--perf-bias, -b
On supported Intel processors, this option sets a register which allows the cpupower utility (or other software/firmware) to set a policy that controls the relative importance of performance versus energy savings to the processor. The range of valid numbers is 0-15, where 0 is maximum performance and 15 is maximum energy efficiency.
The processor uses this information in model-specific ways when it must select trade-offs between performance and energy efficiency. This policy hint does not supersede Processor Performance states (P-states) or CPU Idle power states (C-states), but allows software to have influence where it would otherwise be unable to express a preference.
On SUSE SLES 11 and 12 systems one can set the perf-bias for all CPUs through the cpupower utilit with one of the following commands:
Tuned profiles (Linux)
By using the command line utility tuned-adm, a user may switch between various predefined and user defined tuning profiles. Predefined profiles commonly refer to typical use cases, but one is not limited to using any exact use case on any certain system. Profiles are used to manage different OS features, manage power saving features, and manage hardware options. The default profile is called default and is the lowest predefined profile in regards to powersaving
The throughput-performance profile disabled tuned and ktune power savings features, while enabling sysctl settings that improve through performance of a server's disk, network IO, and switches to the deadline scheduler. This profile also sets the CPU governonr to performance mode
To set a tuned profile, use the command:
Firmware Settings
One or more of the following settings may have been set. If so, the "Platform Notes" section of the report will say so; and you can read below to find out more about what these settings mean.
Intel Hyperthreading Options (Default = Enabled):
This feature allows enabling/disabling of logical processor cores on processors supporting Intel's Hyper-Threading Technology. This option may improve overall performance for applications that will benefit from higher processor core count.
Processor Core Disable (Intel Core Select) (Default = number of physical cores/processor):
This feature allows disabling of processor cores using Intel's Core Multi-Processing (CMP) Technology. This option allows disabling of a specific number of the cores on each physical processor. This option has the following potential uses: Reduce processor power usage and potentially improve performance/watt with some applications; improve overall performance for applications that will benefit from higher performance cores rather than more processing cores; address issues with software that is licensed on a per-core basis.
Specific cores for each socket (for example core0, core1, etc., core14) can be explicitly enabled or disabled. By default all cores of all sockets are enabled.
Memory RAS Configuration (Default = RAS):
This option controls the configuration of memory reliability, availability and serviceability (RAS) features. When setting this option to Maximum Performance system performance is optimized. When setting this option to RAS (also known as lockstep mode) system reliability is enhanced with a degree of memory redundancy.
First created October 19th, 2015.