Best practice workload profile für HPE Proliant Gen10 hardware

[pootch]

Hello.

Would anybody be so kind to review the available workload profiles for the new HPE Proliant Gen10 Hardware.

As for my understanding the profile "high Performance Cumpute" will fit the needs?

Spoiler

Workload Profiles
Workload Profiles and performance options
Workload Profiles is one of the HPE Intelligent System Tuning (IST) features and allows you to tune the
resources in your HPE ProLiant server by choosing a pre-configured workload profile. The server will
automatically configure the BIOS settings to match the selected workload.
System provided Workload Profiles
The system provides these Workload Profiles:
General Power Efficient Compute
This profile is the default profile for most ProLiant servers and HPE Synergy compute modules.
This profile applies the most common performance settings that benefit most application workloads while
also enabling power management settings that have minimal impact to overall performance. The settings
that are applied heavily favor a balanced approach between general application performances versus
power efficiency.
This profile is recommended for customers that do not typically tune their BIOS for their workload.
General Peak Frequency Compute
This profile is intended for workloads that generally benefit from processors or memory that must achieve
the maximum frequency possible, for any individual core, at any time. Power management settings are
applied when they ensure that any component frequency upside can be readily achieved. Processing
speed is favored over any latencies that might occur. This profile is a general-purpose profile, so
optimizations are done generically to increase processor core and memory speed.
This profile benefits workloads that typically benefit from faster compute time.
General Throughput Compute
This profile is intended to be used for workloads where the total maximum sustained workload throughput
is needed. Increased throughput does not always occur when the processor runs at the highest individual
core speed. Increased throughput can occur when the processor is able to perform sustained work across
all available cores during maximum utilization. Power management settings are disabled when they are
known to have impact on maximum achievable bandwidth.
Best throughput is achieved when the workload is also (Non-uniformed Memory Access) NUMA aware
and optimized so settings that benefit NUMA awareness are applied.
Virtualization - Power Efficient
This profile is intended to be used for virtualization environments. The profile ensures that all available
virtualization options are enabled. Certain virtualization technologies can have possible performance
impacts to nonvirtualized environments and can be disabled in other profiles. Power management settings
can have an impact on performance when running virtualization operating systems and this profile applies
power management settings that are virtualization friendly.
Virtualization - Max Performance
This profile is intended to be used for virtualization environments. The profile ensures that all available
virtualization options are enabled. Power management settings are disabled in favor of delivering
maximum performance.
Low Latency
This profile is intended to be used by customers who desire the least amount of computational latency for
their workloads. This profile follows the most common best practices that are documented in the HPE Low
4 Workload Profiles
Latency Whitepaper. Maximum speed and throughput are often sacrificed to lower overall computational
latency. Power management and other management features that might introduce computational latency
are also disabled.
The profile benefits customers running Real-Time Operating Systems (RTOS) or other transactional
latency sensitive workloads.
Mission Critical
This profile is intended to be used by customers who trade off performance for server reliability above the
basic server defaults. The profile enables advanced memory reliability, availability, and serviceability
(RAS) features that are known to have more than a measurable impact to computational performance.
Enabling this profile will have an impact to maximum memory bandwidth and will increase memory
latency.
Transactional Application Processing
This profile is intended to be used for business processing environments, such as online transaction
processing (OLTP) applications that require a database back-end. For example, workloads typically
comprised of a high number of user-based, transactional applications running on a single server with
cohosted database component. The profile balances the requirement of managing both peak frequency
and throughput.
High Performance Compute (HPC)
This profile is intended for customers running in a traditional HPC environment. Typically, these
environments are clustered environments where each node performs at maximum utilization for extended
periods of time to solve large-scale scientific and engineering workloads. The default for our Apollo series
servers, power management is typically disabled in favor of sustained available bandwidth and processor
compute capacity. This profile is similar to the Low Latency profile except that some latency is accepted to
achieve maximum throughput.
Decision Support
This profile is intended for Enterprise Business Database (Business Intelligence) workloads that are
focused on operating and/or accessing data warehouses, such as data mining or online analytical
processing (OLAP).
Graphic Processing
This profile is intended for workloads that are run on server configurations which utilize Graphics
Processing Units (GPUs.) GPUs typically depend on maximum bandwidth between I/O and Memory.
Power management features that have impact on the links between I/O and memory are disabled. Peer to
Peer traffic is also critical and therefore virtualization is also disabled.
I/O Throughput
This profile is intended to be used for configurations that depend on maximum throughput between I/O
and memory. Processor utilization driven power management features that have performance impact to
the links between I/O and memory are disabled.

[Frederic_De_Ranter]

Hi,

It's very interesting to look at the UEFI System Utilities User Guide from HPE where all the options are described in detail (from page 40 describes the workloads and from page 90 it describes the different settings in detail):

https://support.hpe.com/hpsc/doc/public/display?docId=emr_na-a00016407ja_jp

and we have a guideline for generic BIOS settings listed here:

https://community.qlik.com/t5/Qlik-Scalability/Quick-tips-8-Server-Settings-For-Best-Performance/gpm...

If you don't care about power consumption and want an easy setting to put, I think the "High Performance Compute" is the correct one.

If you want to save some power and get a little better performance when only a few cores are being used (higher boost speed for those cores) then you can base your profile on the HPC workload settings, but change the two following settings:

• Minimum Processor Idle Power Core C-state: C6
• Minimum Processor Idle PowerPackage C-state: Package C6 (retention) State

Best regards,

Frederic