Clarifiy sentences in suse-best-practices/xml/MAIN-SBP-AMD-EPYC-3-SLES15SP2.xml

[chabowski]

suse-best-practices/xml/MAIN-SBP-AMD-EPYC-3-SLES15SP2.xml

Line 122 in c08df52

the <quote>distance</quote>. This should be considered when tuning for <emphasis

Hi Dario @dfaggioli
can you clarify the following sentence:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessarily to local memory incurring a variable access penalty.

should it be:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessarily incurring a variable access penalty to local memory.

or:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessary to local memory incurring a variable access penalty.

It is not completely clear currently. Thank you very much!

[chabowski]

@dfaggioli Next sentence to clarify:

"What is not obvious is the links to the IO die, but it should be taken into account when splitting a workload to optimize bandwidth to memory."

"Not obvious are the links to the IO die, but this () should be taken into account .... "
"Not obvious are the links to the IO die, but the IO die should be taken into account ...."

[chabowski]

@dfaggioli and one more:

In addition to lstopo can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache.

Options:

"In addition to lstopo which can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

"In addition, lstopo can provide the information, the level, size and ID of CPUs that share cache, which can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

[dfaggioli]

suse-best-practices/xml/MAIN-SBP-AMD-EPYC-3-SLES15SP2.xml

Line 122 in c08df52

the <quote>distance</quote>. This should be considered when tuning for <emphasis

Hi Dario @dfaggioli
can you clarify the following sentence:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessarily to local memory incurring a variable access penalty.

I think it means that it can happen that a memory operation accesses some remote (i.e., non-local) memory. That is, memory accesses not necessarily always happen to local memory, which mean they can incur a variable access penalty.

So, this, I would say:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessary to local memory incurring a variable access penalty.

However, can we double check with @gormanm , as I think this comes from him ?

[dfaggioli]

@dfaggioli Next sentence to clarify:

"What is not obvious is the links to the IO die, but it should be taken into account when splitting a workload to optimize bandwidth to memory."

This sentence is right below the figure fig-epyc-topology. I think it means that it is not obvious, by looking at that figure, how the various CCXs/dies (which are shown in the figure itself) are connected to the IO die (which is not present in the figure because the tool does not draw it).

But again, @gormanm would be the one source of truth here. :-)

[dfaggioli]

@dfaggioli and one more:

In addition to lstopo can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache.

Options:

"In addition to lstopo which can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

"In addition, lstopo can provide the information, the level, size and ID of CPUs that share cache, which can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

This is still from @gormanm . Personally, I'd go for something like:

"In addition to the information that lstopo can provide, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

But we should really double check with him (plus, I'm no native Eng. speaker, so don't trust me!)

[gormanm]

suse-best-practices/xml/MAIN-SBP-AMD-EPYC-3-SLES15SP2.xml

Line 122 in c08df52

the <quote>distance</quote>. This should be considered when tuning for <emphasis

Hi Dario @dfaggioli
can you clarify the following sentence:
This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessarily to local memory incurring a variable access penalty.

I think it means that it can happen that a memory operation accesses some remote (i.e., non-local) memory. That is, memory accesses not necessarily always happen to local memory, which mean they can incur a variable access penalty.

So, this, I would say:

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses are not necessary to local memory incurring a variable access penalty.

However, can we double check with @gormanm , as I think this comes from him ?

The switch from "necessarily" to "necessary" changes the semantics of the sentence. Alternative suggestion as follows

This should be considered when tuning for Non-Uniform Memory Architecture (NUMA) as all memory accesses may not reference local memory incurring a variable access penalty.

[gormanm]

@dfaggioli and one more:

In addition to lstopo can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache.

Options:

"In addition to lstopo which can provide the information, the level, size and ID of CPUs that share cache can be identified from the files under /sys/devices/system/cpu/cpuN/cache."

This version is the best.

[gormanm]

@dfaggioli Next sentence to clarify:

"What is not obvious is the links to the IO die, but it should be taken into account when splitting a workload to optimize bandwidth to memory."

"Not obvious are the links to the IO die, but the IO die should be taken into account ...."

This option is the better of the two. Thanks!