Enhancement: Move pg_profile_lookup.ini to a shared place for all HWSKU on a specific platform

[bingwang-ms]

Is it platform specific

generic

Importance or Severity

Critical

Description of the enhancement

The enhancement is to move pg_profile_lookup.ini to a shared place for all HWSKUs on a specific platform so that we don't need to define this lookup table for each HWSKU.

Current Behavior

There is pg_profile_lookup.ini in each HWSKU folder, for example

For HWSKU Arista-7060X6-64PE-B-P32O64

sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-P32O64/pg_profile_lookup.ini

Lines 1 to 8 in 587e725

	# PG lossless profiles.
	# speed cable size xon xoff threshold xon_offset
	400000 30m 8636 0 425958 0 3556
	400000 300m 8636 0 839216 0 3556
	400000 500m 8636 0 1136396 0 3556
	100000 500m 8636 0 1136396 0 3556
	800000 5m 8636 0 724662 0 3556
	800000 300m 8636 0 1621536 0 3556

For HWSKU Arista-7060X6-64PE-B-P64

sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-P64/pg_profile_lookup.ini

Lines 1 to 3 in 587e725

	# PG lossless profiles.
	# speed cable size xon xoff threshold xon_offset
	800000 5m 8636 0 724662 0 3556

Proposed Behavior

Share one common pg_profile_lookup.ini for each HWSKU. The file can be put in platform folder or default_hwsku folder

[r12f]

although this issue solves the pg_profile_lookup.ini, but it won't solve the qos template difference between the HWSKUs, so we also need to have that covered too.

e.g. Arista-7060X6-64PE-B-C512S2 vs Arista-7060X6-64PE-B-P64

[bingwang-ms]

although this issue solves the pg_profile_lookup.ini, but it won't solve the qos template difference between the HWSKUs, so we also need to have that covered too.

e.g. Arista-7060X6-64PE-B-C512S2 vs Arista-7060X6-64PE-B-P64

That's correct. The QoS template enhancement will be tracked with another enhancement since it's HWSKU dependent for some platform, while pg_profile_lookup.ini can always be shared among different HWSKU.

[Bojun-Feng]

Seems like the simplest approach is to add a platform-level fallback to buffermgrd.sh — check /usr/share/sonic/hwsku/pg_profile_lookup.ini first, and if it's not there, fall back to /usr/share/sonic/platform/pg_profile_lookup.ini. The platform directory is already mounted into all containers, so no infrastructure changes should be needed.

[Bojun-Feng]

pg_profile_lookup.ini can always be shared among different HWSKU.

This isn't always the case. I looked at the codebase and some platforms have HWSKUs with genuinely different buffer tuning for the same (speed, cable_length) key.

For example, on x86_64-arista_7060x6_64pe_b, these two files have rows that differ only in xon_offset:

sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-O128/pg_profile_lookup.ini

Line 8 in f56312f

100000 500m 8636 0 1136396 0 3356

sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-P32O64/pg_profile_lookup.ini

Line 6 in f56312f

100000 500m 8636 0 1136396 0 3556

This pattern repeats in several other platforms that have non-identical pg_profile_lookup.ini files (Arista, Mellanox/Nvidia, Marvell).

Any solution here would still need a per-HWSKU override mechanism — e.g., HWSKU-level file takes precedence, with platform-level as fallback for platforms where all HWSKUs share the same values.

[bingwang-ms]

pg_profile_lookup.ini can always be shared among different HWSKU.

This isn't always the case. I looked at the codebase and some platforms have HWSKUs with genuinely different buffer tuning for the same (speed, cable_length) key.

For example, on x86_64-arista_7060x6_64pe_b, these two files have rows that differ only in xon_offset:

sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-O128/pg_profile_lookup.ini

Line 8 in f56312f

100000 500m 8636 0 1136396 0 3356
sonic-buildimage/device/arista/x86_64-arista_7060x6_64pe_b/Arista-7060X6-64PE-B-P32O64/pg_profile_lookup.ini

Line 6 in f56312f

100000 500m 8636 0 1136396 0 3556
This pattern repeats in several other platforms that have non-identical pg_profile_lookup.ini files (Arista, Mellanox/Nvidia, Marvell).

Any solution here would still need a per-HWSKU override mechanism — e.g., HWSKU-level file takes precedence, with platform-level as fallback for platforms where all HWSKUs share the same values.

That's a good catch. Per my understanding, it supposed to be the same value on the same platform for the same cable_length and speed combination. @dakotac-arista Can you please double check on this?

[bingwang-ms]

Cross-reference related issues Part of the broader Generic HWSKU effort:
#25718 (hwsku.json),
#25747 (buffer/QoS templates),
#25720 (SAI profile),
#24494 (port_config.ini).