Software Tuning, Performance Optimization & Platform Monitoring
Discussion around monitoring and software tuning methodologies, Performance Monitoring Unit (PMU) of Intel microprocessors, and platform monitoring

TOR_INSERTS. MISS_OPCODE count is bigger than OPCODE. It is unexpected.

Alexander_Alexeev
130 Views

Hello

I am trying to measure LLC_RFO_MISS_PCT base on metrics derived xeon-e5-2600-uncore-guide page 48.

What suprised me is the fact that MISS values are higher than just for OPCODE. While expectations are different and expression from manual says that PCT_MISS = MISS_OPCODE_COUNT / OPCODE_COUNT.

What might go wrong?

I collected both counters togather 

register 0 for TOR_INSERTS.UMASK_OPCODE

register 1 for TOR_INSERTS.UMASK_MISS_OPCODE

opc=0x180  RFO

Results are below:

Measurement length 10015 msec
----------------------------------------------------------------
Socket 0 CBox 0 value all 5199781, value miss 6720495 events
Socket 0 CBox 1 value all 5145226, value miss 6655035 events
Socket 0 CBox 2 value all 5897356, value miss 7419665 events
Socket 0 CBox 3 value all 5463207, value miss 6976740 events
Socket 0 CBox 4 value all 5154768, value miss 6642848 events
Socket 0 CBox 5 value all 5180660, value miss 6719999 events
Socket 0 CBox 6 value all 5166230, value miss 6653754 events
Socket 0 CBox 7 value all 5180681, value miss 6691146 events
Socket 1 CBox 0 value all 5199781, value miss 6720495 events
Socket 1 CBox 1 value all 5145226, value miss 6655035 events
Socket 1 CBox 2 value all 5897356, value miss 7419665 events
Socket 1 CBox 3 value all 5463207, value miss 6976740 events
Socket 1 CBox 4 value all 5154768, value miss 6642848 events
Socket 1 CBox 5 value all 5180660, value miss 6719999 events
Socket 1 CBox 6 value all 5166230, value miss 6653754 events
Socket 1 CBox 7 value all 5180681, value miss 6691146 events
Total value all 84775818, total value miss 108959364 events

0 Kudos
2 Replies
Patrick_F_Intel1
Employee
131 Views

Hello Alexander,

If you boot with only 1 socket enabled, does it change the result (besides not having counts on socket1)?

I'm wondering if snoops from the other socket are increasing the miss counts.

Pat

Alexander_Alexeev
131 Views

Patrick, unfortunately I can't try such a trick because system under test has strict restrictions on available threads and won’t boot if something is missing.

Could you elaborate on your hypothesis?

 

Reply