search

EEHPCWG / PowerMeasurementMethodology

TeX for PMM spec
9 stars 1 forks source link

System boundary definition: cooling #106

Open tilsche opened 1 year ago

tilsche commented 1 year ago

Problem statement

For cooling and subsystems, it seems difficult to narrow down and extract the range corresponding to the measurement target. Especially when the system is large scale or when multiple systems are installed.

Solution

Try to find a way to be more precise with the system boundary definition with respect to cooling. Ideally, find widely-used definitions that we can reference in other standard documents.

Can also be aided by examples #104 .

The challenging aspect here will be to balance fairness and practicality. Currently we require any internal cooling devices (self-contained liquid cooling systems and fans).. This could be seen as an unfair advantage of systems with (high) external cooling costs. On the other hand, it is hard to separate such costs for measurement.

tilsche commented 1 year ago

As per today's meeting, Ian and Eric are currently working on this. It turns out to be tightly integrated with #107

chriswasser commented 2 months ago

Additional information based on discussion after the RWTH Green500 presentation on 2024-07-16: The portion of energy consumed by the rack-level side coolers and CDUs was measured to be 0.5-1.5% of the energy consumption of the compute nodes (during normal operation not just during the Green500 run). Therefore, including these components as internal cooling devices only effects the reported efficiency in a minor fashion.