Skip to content

Latest commit

 

History

History
113 lines (79 loc) · 3.53 KB

monitoring.md

File metadata and controls

113 lines (79 loc) · 3.53 KB
Raw

Monitoring

hwbench is using a specific monitoring engine to collect data from several sources :

  • BMC
  • PDU
  • Turbostat

Regarding the source, power, frequencies, thermal or cooling metrics are collected.

Concept

During each benchmark, and if the monitoring is enabled, metrics are collected every 2 seconds and aggregated every 10 seconds to get statistics over this period of time.

At the end of the benchmark the monitoring metrics are added in the a result file. hwgraph will use them to plot how these components behave during the benchmark.

Usage

To enable the monitoring feature, just set the monitor directive in the configuration file.

As per this release, only monitor=all is supported. In future releases, it will be possible to list the source to be considered.

Configuration file

When monitoring is enabled, the -m <config_file> option must be used to describe the server's configuration. This file is separated from the job file as it could be specific to each host.

Each source is defined like the following :

  [section_name]
  username=<username>
  password=<password>
  type=<type>
  url=<url>

BMC

When defining a BMC, the type must be set to BMC. The section_name could the hardware vendor name like DELL, HPE, or default.

A typical example looks like :

[HPE]
username=Administrator
password=YOURPASSWORD
type=BMC

Note: if no url parameter is provided, it will be automatically detected at runtime via ipmitool tool (or ilorest on HPE systems).

A single BMC configuration will be used per server and vendor specific will be selected first if matching the running hardware.

The BMC code is using redfish endpoints to monitor the server. Vendor specific endpoints can be used in addition of the generic ones to get all meaningful metrics.

Hwbench monitoring code requires the BMC to be reachable from the host.

PDU

When defining a PDU,

  • a section_name, a user-defined value to represent this PDU
  • the type must be set to PDU.
  • a driver must be chosen
  • an URL is required
  • the outlet port must be selected

A typical example looks like :

[myPDU]
username=admin
password=admin
type=PDU
driver=raritan
url=http://mypdu/
outlet=21

Note: Several PDU configurations can be defined and used simultaneously.

Driver

There exist many PDU providers and the software quality may vary a lot and so the protocols. To ensure a good compatibility with them, drivers can be added to hwbench.

For this release, only raritan driver exists but as it uses some redfish endpoints, it might work on other products.

If you have tested it on some other PDUs or have created a custom driver, feel free to push a PR for review.

Note: The Raritan driver only exports the power in Watts but can be expanded easily to get more metrics.

URL

The url cannot be automatically detected so it must be provided to hwbench.

Outlet

This directive selects the physical outlet where the server is connected.

Outletgroup

Some products support outlet groups where outlets from different PDUs are grouped in a single outletgroup.

If the PDU supports it, the outletgroup can be used to specify which one to use. A typical example looks like :

[PDU_with_grouped_outlets]
username=admin
password=admin
type=PDU
driver=raritan
url=https://mypdu/
outletgroup=1

Note: outlet and ``outletgroup` are mutually exclusive.

Turbostat

Turbostat will be automatically used on x86_64 systems if already installed on the server with release >= 2022.04.16. No configuration is required.