L3CAT not giving the expected results on the caterpillar benchmark

[apinas]

Hi!
I was trying to reproduce the Intel ECI caterpillar results by running the benchmark with and without L3 Caché allocation on a i7-13700E machine.
I am using kernel 6.12.11 with PREEMPT_RT.

Basically, I have some isolated cores in which I run the application, while a non-isolated core creates some "noise" by using memcpy:

[apinas@zodd ~]$ cat /sys/devices/system/cpu/isolated
4-7

Using the pqos tool I partition the L3 cache so the isolated cores get half of the cache, and the non-isolated cores get the remaining. To do this, I run:

export RDT_PROBE_MSR=1
pqos -R
pqos -e "llc:0=0x03f;llc:3=0xfc0"
pqos -a "llc:0=0-3,8-15;llc:3=4-7"

The output is

sudo -E pqos -R
NOTE: Mixed use of MSR and kernel interfaces to manage
CAT or CMT & MBM may lead to unexpected behavior.
WARN: CPUID.0x7.0: Monitoring capability not supported!
WARN: Cache allocation not supported on model name '13th Gen Intel(R) Core(TM) i7-13700E'!
ERROR: RDMSR failed for reg[0xca0] on lcore 0
WARN: Cache allocation not supported on model name '13th Gen Intel(R) Core(TM) i7-13700E'!
ERROR: RDMSR failed for reg[0xca0] on lcore 0
Allocation reset successful

sudo -E pqos -e "llc:0=0x03f;llc:3=0xfc0"
NOTE: Mixed use of MSR and kernel interfaces to manage
CAT or CMT & MBM may lead to unexpected behavior.
WARN: CPUID.0x7.0: Monitoring capability not supported!
WARN: Cache allocation not supported on model name '13th Gen Intel(R) Core(TM) i7-13700E'!
ERROR: RDMSR failed for reg[0xca0] on lcore 0
SOCKET 0 L3CA COS0 => MASK 0x3f
SOCKET 0 L3CA COS3 => MASK 0xfc0
Allocation configuration altered.

[apinas@zodd intel]$ sudo -E pqos -a "llc:0=0-3,8-15;llc:3=4-7"
NOTE: Mixed use of MSR and kernel interfaces to manage
CAT or CMT & MBM may lead to unexpected behavior.
WARN: CPUID.0x7.0: Monitoring capability not supported!
WARN: Cache allocation not supported on model name '13th Gen Intel(R) Core(TM) i7-13700E'!
ERROR: RDMSR failed for reg[0xca0] on lcore 0
Allocation configuration altered.

I can tell that the configuration has been applied:

sudo -E pqos -V -s
NOTE: Mixed use of MSR and kernel interfaces to manage
CAT or CMT & MBM may lead to unexpected behavior.
INFO: Requested interface: AUTO
INFO: resctrl not detected. Kernel version 4.10 or higher required
INFO: Selected interface: MSR
INFO: CACHE: type 1, level 1, max id sharing this cache 2 (1 bits)
DEBUG: CACHE: not inclusive, direct mapped, 8 way(s), 64 set(s), line size 64, 1 partition(s)
INFO: CACHE: type 2, level 1, max id sharing this cache 2 (1 bits)
DEBUG: CACHE: not inclusive, direct mapped, 8 way(s), 128 set(s), line size 64, 1 partition(s)
INFO: CACHE: type 3, level 2, max id sharing this cache 8 (3 bits)
DEBUG: CACHE: not inclusive, direct mapped, 16 way(s), 4096 set(s), line size 64, 1 partition(s)
INFO: CACHE: type 3, level 3, max id sharing this cache 128 (7 bits)
DEBUG: CACHE: not inclusive, complex cache indexing, 12 way(s), 40960 set(s), line size 64, 1 partition(s)
DEBUG: Detected core 0, socket 0, NUMAnode 0, L2 ID 0, L3 ID 0, APICID 0
DEBUG: Detected core 1, socket 0, NUMAnode 0, L2 ID 1, L3 ID 0, APICID 8
DEBUG: Detected core 2, socket 0, NUMAnode 0, L2 ID 2, L3 ID 0, APICID 16
DEBUG: Detected core 3, socket 0, NUMAnode 0, L2 ID 3, L3 ID 0, APICID 24
DEBUG: Detected core 4, socket 0, NUMAnode 0, L2 ID 4, L3 ID 0, APICID 32
DEBUG: Detected core 5, socket 0, NUMAnode 0, L2 ID 5, L3 ID 0, APICID 40
DEBUG: Detected core 6, socket 0, NUMAnode 0, L2 ID 6, L3 ID 0, APICID 48
DEBUG: Detected core 7, socket 0, NUMAnode 0, L2 ID 7, L3 ID 0, APICID 56
DEBUG: Detected core 8, socket 0, NUMAnode 0, L2 ID 8, L3 ID 0, APICID 64
DEBUG: Detected core 9, socket 0, NUMAnode 0, L2 ID 8, L3 ID 0, APICID 66
DEBUG: Detected core 10, socket 0, NUMAnode 0, L2 ID 8, L3 ID 0, APICID 68
DEBUG: Detected core 11, socket 0, NUMAnode 0, L2 ID 8, L3 ID 0, APICID 70
DEBUG: Detected core 12, socket 0, NUMAnode 0, L2 ID 9, L3 ID 0, APICID 72
DEBUG: Detected core 13, socket 0, NUMAnode 0, L2 ID 9, L3 ID 0, APICID 74
DEBUG: Detected core 14, socket 0, NUMAnode 0, L2 ID 9, L3 ID 0, APICID 76
DEBUG: Detected core 15, socket 0, NUMAnode 0, L2 ID 9, L3 ID 0, APICID 78
WARN: CPUID.0x7.0: Monitoring capability not supported!
INFO: Monitoring capability not detected
INFO: CPUID.0x7.0: L3 CAT not detected. Checking brand string...
DEBUG: CPU brand string '13th Gen Intel(R) Core(TM) i7-13700E'
WARN: Cache allocation not supported on model name '13th Gen Intel(R) Core(TM) i7-13700E'!
INFO: Checking model and family ID...
INFO: Probing msr....
ERROR: RDMSR failed for reg[0xca0] on lcore 0
INFO: L3CA capability detected
INFO: L3 CAT details: CDP support=0, CDP on=0, #COS=16, #ways=12, ways contention bit-mask 0x0
INFO: L3 CAT details: cache size 31457280 bytes, way size 2621440 bytes
INFO: L3 CAT details: I/O RDT support=0, I/O RDT on=0
INFO: CPUID.0x7.0: L2 CAT not supported
INFO: L2CA capability not detected
INFO: CPUID.0x7.0: MBA not supported
INFO: MBA capability not detected
DEBUG: allocation init OK
DEBUG: monitoring init aborted: feature not present
DEBUG: I/O RDT init aborted: feature not present
L3CA COS definitions for Socket 0:
L3CA COS0 => MASK 0x3f
L3CA COS1 => MASK 0xfff
L3CA COS2 => MASK 0xfff
L3CA COS3 => MASK 0xfc0
L3CA COS4 => MASK 0xfff
L3CA COS5 => MASK 0xfff
L3CA COS6 => MASK 0xfff
L3CA COS7 => MASK 0xfff
L3CA COS8 => MASK 0xfff
L3CA COS9 => MASK 0xfff
L3CA COS10 => MASK 0xfff
L3CA COS11 => MASK 0xfff
L3CA COS12 => MASK 0xfff
L3CA COS13 => MASK 0xfff
L3CA COS14 => MASK 0xfff
L3CA COS15 => MASK 0xfff
Core information for socket 0:
Core 0, L2ID 0, L3ID 0 => COS0
Core 1, L2ID 1, L3ID 0 => COS0
Core 2, L2ID 2, L3ID 0 => COS0
Core 3, L2ID 3, L3ID 0 => COS0
Core 4, L2ID 4, L3ID 0 => COS3
Core 5, L2ID 5, L3ID 0 => COS3
Core 6, L2ID 6, L3ID 0 => COS3
Core 7, L2ID 7, L3ID 0 => COS3
Core 8, L2ID 8, L3ID 0 => COS0
Core 9, L2ID 8, L3ID 0 => COS0
Core 10, L2ID 8, L3ID 0 => COS0
Core 11, L2ID 8, L3ID 0 => COS0
Core 12, L2ID 9, L3ID 0 => COS0
Core 13, L2ID 9, L3ID 0 => COS0
Core 14, L2ID 9, L3ID 0 => COS0
Core 15, L2ID 9, L3ID 0 => COS0

Then I run the caterpillar test as:

/opt/benchmarking/caterpillar/start-benchmark.py --irq_core 0 --nn_core 3 --caterpillar_args '-c 6 -l results.txt -s 5000'

After running for some hours, with and without CAT enabled I got the following results:

There does not seem to be much of a difference between the CAT-enabled and the non-CAT-enabled runs (in fact, the CAT version reports more jitter), contrary to what happens in the results shown on the Intel ECI site.

I hope you could help me understand what is happening, and how I could get better results!

As a side note, I assume the pqos utility is using the MSR interface, because despite having enabled the CONFIG_X86_CPU_RESCTRL=y in the kernel, and set the following kernel argument rdt=l3cat, I am unable to mount the resctrl filesystem:

mount -t resctrl resctrl /sys/fs/resctrl
mount: /sys/fs/resctrl: mount point does not exist.

[rkanagar]

Hi @apinas

Refer resctrl kernel documentation.
https://docs.kernel.org/arch/x86/resctrl.html

mount -t resctrl resctrl [-o cdp[,cdpl2][,mba_MBps][,debug]] /sys/fs/resctrl

I can see allocation & monitoring features are not supported in your machine.

Please check them using below commands:
pqos --iface=msr -d
pqos --iface=msr -D

cpuid -1 -l 0x10
cpuid -1 -l 0x10 -s 2
cpuid -1 -l 0x10 -s 1

Thanks,
Raghavan K.

[apinas]

Hi @rkanagar,
by running the commands you provided I can see that the cache allocation is not supported because it does not support RDT. However, as reported by the Intel® Resource Director Technology (Intel® RDT) Architecture Specification, Table 7-9, the Intel Core i7-13700E Processor supports L3 cache allocation.
In the previously mentioned page Intel ECI caterpillar results, the testbench is ran for an Intel Core i7-1185GRE Processor. This processor does not support RDT, but as well as the i7-13700E it implements a "non-architectural version" of L3 cache allocation. I think I should be able to obtain the same results.