Cisco UCS – BIOS settings – part 3

Processor – basic options

Execute Disabled Bit:
This option will prevent malicious code from bufferflow attacks by classifying memory areas were code can be executed.
Depending on the support of  operating system you should Enable this option.

Direct Cache Access:
This option allows data from I/O devices to be placed directly in the processor cache increasing I/O performance.
You should always Enable this setting.

Local X2 Application Policy Infrastructure Controller (APIC): << WRONG NAME!!
it should be “local x2APIC” (standing for x2 Advanced Programming Interrupt controller)!
The APIC controls the interrupts on a microprocessor, newer versions of this APIC were developed by extending the addressability of the processors.”Local x2APIC” is such kind of controller.
My recommendation is to set this option to Auto.

Frequancy Floor Override:
This option controls the way may (or may not) a processor may drop below it’s maximum non-turbo frequency (essentially disabling P-states). By enabling this option you may improve system performance but at the sacrificies of power consumption.
My recommendation is to set this option to Disable for general workloads, only for VDI you would Enable this settings.

P-STATE Coordination:
This option defines how the P-states will be supported. P-states (through ACPI) are states of a processor were they can lower the frequency and reducing power consumption. The higher the Pn state the lower the frequency beecomes and the less power the processor will consume.
For more information click here (Intel blog)

P-states don’t have an impact on performance because it adjust its state to the utilization of a processor.
The P-states can be controlled by the hardware (HW_ALL) or by the OS (SW_ANY or SW_ALL). Generally you would let the OS decide how the P-states should be used. The OS must support ACPI (Windows and ESXi support ACPI).
My recommendation is to let the Power Management setting control this setting.

In VMware you could set how the P-states should be used by adjusting the Power Management setting:
vspherepowerpolicies
With I/O latency sensitive workloads you can consider to High Performance setting, which will prevent P-states and C-states beyond C1. One downside with this setting  is that Intel Turbo Boost is disabled because this technology relies on P- and C-states (which are disabled). USE WITH CAUTION!

DRAM Clock Throttling:
As with a processor the same applies to DRAM modules: the higher the frequency the higher the power consumption.
By setting this option to Auto you’re ginving the control to the CPU.
By setting this option to Energy Efficient (low frequency),  Balanced (medium) , Performance (fast) you manually control the frequency.
My recommendation is to set this option to Performance, the power savings from lowering the frequency for the memory modules in minimal.

Channel interleaving:
Channel interleaving divides memory blocks and spreads contiguous portions of data across interleaved channels to enable simultaneous read operation, which results in beter I/O througput.
My recommendation is to set this option to Auto, the system will set the correct interleave option.

Rank Interleaving:
Rank interleaving divides physical ranks of memory so that one rank can be accessed while another is being refreshed.
My recommendation is to set this option to Auto, the system will set the correct interleave option.

Altitude:
Because cooling-fans have less impact at higher altitude, you can configure the approximate number of meters above sea level at which the physical server is installed. The cooling fans will be adjusted to the the configured altitiude.
My recommendation is to set this option to Auto, the thermal sensor should compensate the impact of the altitude automatically.

CPU Hardware Power Management:
With the Intel Broadwell (v4) processors a technique called Hardware Power Management (HWPM) is introduced, which is an additionel process what sits between the OS and the P- and C-states. It is able to make optimal decisions within milliseconds (instead of seconds).
You can disabble the HWPM technique and use the old-style (OS) power management -or- you could use the HWPM in a native-mode were the OS gets an interface or an OOB-mode where the HWPM makes it own decisions IF you have Intel Broadwell (v4) processors.

When you use HWPM-native- or HWPM-OOB-mode the options Intel Speedstep and Intel Turboboost are being ignored, they are being controlled by the HWPM module.
My recommendation is to set this option to HWPM-native-mode, which gives a optimal mix between performance,  power consumption.and control by the OS.

More information can be found here

Prefetcher options

Prefetching is a technique were data is proactivly placed in the L2 cache of the processor were it is pulled from the memory before the processor actually request the data. This could improve performance because cache misses are being eliminated. A downside is the increased bandwidth utilization on the processor-bus.
dependent on the application prefetching could give a positive or a negative result.
You always should test prefetching with you application/environment.

More information can be found here

Hardware Prefetcher
Without programmer intervention the hardware itself makes decisions wich data must be proactivly placed in the L2 cache.
My recommendation is to let the Power Management setting control this setting.

Adjacent Cache Line Prefetcher
Adjacent Cache-Line Prefetch works just like Hardware Prefetcher without programmer intervention. When a processor request for a 64 bit cache line an additional 64 bit is being cached, so when the processor requests this extra cache line it’s already available and no cache miss occures.
My recommendation is to let the Power Management setting control this setting.

Data Cache Unit (DCU) Streamer Prefetcher
This option controls the L1 cache of the processors by detecting multiple loads from the same cache line within a time limit, in order to then prefetch the next line from the L2 cache or the main memory into the L1 cache based on the assumption that the next cache line will also be needed.
My recommendation is to let the Power Management setting control this setting.

Data Cache Unit (DCU) IP Prefetcher
This option controls the L1 cache of the processors by detecting previous sequential accesses and attempts on this basis to determine the next data to be expected and, if necessary, to prefetch this data from the L2 cache or the main memory into the L1 cache.
My recommendation is to let the Power Management setting control this setting.

Device Type options

Intel Turbo Boost
Intel Turbo Boost accelerates processor and graphics performance for peak loads automatically allowing processor cores to run faster than the rated operating frequency.
Turbo Boost is by default Enabled, but it depends on EIST, which itselfs depends on the C1E-state. If you disable EIST or the C1E state, Intel Turbo Boost is disabled.
My recommendation is to let the Power Management setting control this setting.
The Power Management option must be set to Custom otherwise this setting is ignored!

More information can be found here.

Enhanced Intel Speedstep (EIST)
EIST is a technology whereby the frequency of processor cores can be lowered (through P-states) by software to reduce power consumption. It works in conjunction with P-states.
When the ability of using EIST is disabled, Turbo Boost is also disabled!
My recommendation is to let the Power Management setting control this setting.
For VDI environments whereby the changes between P-state give a small delay in the desktop experience: my recommendation is to Disable EIST for VDI/TS.

The Power Management option must be set to Custom otherwise this setting is ignored!

More information can be found here.

Hyper Threading (HT)
HT is a technology whereby a processor can work on 2 threads simultaneous, which can gain about 30% faster processor performance per core. When a core is waiting (for example  a cache action) for a primairy thread, the same core can work on a secondary thread. This allows a core to work on 2 threads simultaneous. A downside could be that a core is working on the second thread and that the primairy thread must wait until the core is finished on the second thread.

My recommendation is to set this option to Enable HT, because it gives a 30% performance boost. For a very small amount of workloads HT slows down the processing performance. Those workloads mostly have an explicit mandate stating HT should be Disabled.

More information can be found here.

Core Multi-Processing
Every server processor today has multiple cores, with this option you can set the amount of active cores. You can limit the amount of cores for licensing purposes, whereby every core is licensed.
My recommendation is to set this option to All.

Virtualization Technology (VT)
A technology whereby a virtual machine can use the processer without any compatibility issues and performance hits.Intel VT is required for vSphere.
You should always enable Intel-VT.
You must reboot the server before this setting takes effect.

More infromation could be found here.

Power options

Althrough there is already talked a lot about power settings, the settings on this page give direct control over a couple of the previous settings.
The diagrom below shows how  power settings interact with each other.bios-power-settings

The C- and P-states are directly regulating the performance and power of the processors and cores. How the C- and P-states are operating is dependend on various (global) settings which will be discussed below:

Power Management.
This setting have a direct control of EIST, Turbo Boost, C6-state and the Energy Performance settings.
The Custom setting provide a way were you can individually configure EIST, Turbo Boost, C6 state and Energy Performance settings.
All other options are a preconfigured settings (for EIST, TB, C6 and EP).
My recommendation is to set this option to Performance.

Energy Performance.
This setting controls the way the BIOS interacts with the Power Control Unit (PCU) by depending on the information provided by the OS. You can influence it’s decisions for more performance or more power efficiency. The Power Management option must be set to Custom otherwise this individual setting is ignored!
My recommendation is to use the Power Management setting

Processor C state
This settings controls if  the BIOS may or may not use the C-states.
Take caution when you disable the C-state entirely: you also disable the C1E state, which is a no impact C-state but delivers power savings. You should rather use the setting Package C State Limit for disabling (limiting) the C-states process and gain performance.
My recommendation is to set this option to Enable.

Processor C1E
The C1E state is a combination of a C1 state (CPU Idle but powered on) and a P-state (which lowers de processor frequency) without performance impact: There is no delay when the C1E state transfers to a C0 state (full CPU utilization).
My recommendation is to set this option to Enable.

CPU performance
This setting control the prefetchers settings and is NOT dependend on other settings.
My recommendation is to set this option to Enterprise for vSphere/VDI environments.

Package C State Limit
This setting controls which C-state may be used. The higher a C-state, the lower the power consumption but it takes longer for a CPU to become active again.
because C-states beyond C1E give a performance degradation my recommendation is to use the C1 State setting. This limits the C-states to C1(e) which has little to no performance impact, but provides power savings.

Errors and Reporting options

Processor C3 Report
Processor C6 Report
Processor C7 Report
These settings provide the OS information through ACPI C-state reports, when you don’t use these C-states (by limiting the C-states with the Package C State Limit option) you can Disable all of these reports.

Max Variable MTRR Setting
The memory type range registers (MTRRs) provide a mechanism for associating the memory types with physical address ranges in system memory. They allow the processor to optimize operations for different types of memory such as RAM, ROM, frame-buffer memory and memory-mapped I/O devices.
MTRRs in a processor supporting Intel Hyper-Threading Technology are shared by logical processors. When one logical processor updates the setting of the MTRRs, settings are automatically shared with the other logical processors in the same physical package.
My recommendation is to set this option to Auto-Max, to maximize the variable MTTR settings.

More information can be found here (par. 11.11)

Demand Scrub
When a memory single bit error is detected by the CPU, this option will (or will) not correct issues (if it’s possible).
My recommendation is to set this option to Enable.

Patrol Scrub
This option allows the system to periodically scan and correct (single bit) memory errors.
My recommendation is to set this option to Enable.

2 Comments

  1. […] Cisco UCS – BIOS settings – part 2 Cisco UCS – BIOS settings – part 3 […]

    Reply
  2. […] VorigeVolgende […]

    Reply

Leave a Reply

Your email address will not be published. Required fields are marked *

Scroll to top