ClusterCockpit/cc-examples
1
0
Fork 0
mirror of https://github.com/ClusterCockpit/cc-examples.git synced 2025-07-27 23:26:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

FAU config

Browse Source
This commit is contained in:
Jan Eitzinger
2023-06-13 07:26:59 +02:00
parent 19ad3e6953
commit 39c152b71f
58 changed files with 5946 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

145
fau-systems/cc-metric-collector/alex/collectors.json Normal file
View File

@@ -0,0 +1,145 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"nvidia" : {
"use_pci_info_as_type_id": true
},
"lustrestat" : {
"send_all_metrics" : true,
"use_sudo": false,
"send_diff_values": true,
"send_derived_values": true,
"send_abs_values": false
},
"netstat" : {
"include_devices" : [
"enp1s0",
"enp70s0f0",
"enp195s0f0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"ibstat" : {
"send_abs_values": true,
"send_derived_values": true
},
"ipmistat" : {
"send_abs_values": true,
"send_derived_values": true
},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC1": "ACTUAL_CPU_CLOCK",
"FIXC2": "MAX_CPU_CLOCK",
"PMC0": "RETIRED_INSTRUCTIONS",
"PMC1": "CPU_CLOCKS_UNHALTED",
"PMC2": "RETIRED_SSE_AVX_FLOPS_ALL",
"PMC3": "MERGE",
"DFC0": "DRAM_CHANNEL_0",
"DFC1": "DRAM_CHANNEL_1",
"DFC2": "DRAM_CHANNEL_2",
"DFC3": "DRAM_CHANNEL_3"
},
"metrics": [
{
"name": "ipc",
"calc": "PMC0/PMC1",
"type": "hwthread",
"publish": true
},
{
"name": "flops_any",
"calc": "1E-9*PMC2/time",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"name": "clock",
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"type": "hwthread",
"unit": "MHz",
"publish": true
},
{
"name": "mem1",
"calc": "1E-9*(DFC0+DFC1+DFC2+DFC3)*64.0/time",
"unit": "Gbyte/s",
"type": "socket",
"publish": false
}
]
},
{
"events": {
"DFC0": "DRAM_CHANNEL_4",
"DFC1": "DRAM_CHANNEL_5",
"DFC2": "DRAM_CHANNEL_6",
"DFC3": "DRAM_CHANNEL_7",
"PWR0": "RAPL_CORE_ENERGY",
"PWR1": "RAPL_PKG_ENERGY"
},
"metrics": [
{
"name": "core_power",
"calc": "PWR0/time",
"unit": "Watt",
"type": "hwthread",
"publish": true
},
{
"name": "cpu_power",
"calc": "PWR1/time",
"type": "socket",
"unit": "Watt",
"publish": true
},
{
"name": "mem2",
"calc": "1E-9*(DFC0+DFC1+DFC2+DFC3)*64.0/time",
"unit": "Gbyte/s",
"type": "socket",
"publish": false
}
]
}
],
"globalmetrics": [
{
"name": "mem_bw",
"calc": "mem1+mem2",
"type": "socket",
"unit": "Gbyte/s",
"publish": true
}
]
}
}

8
fau-systems/cc-metric-collector/alex/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/alex/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

58
fau-systems/cc-metric-collector/alex/router.json Normal file
View File

@@ -0,0 +1,58 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "alex",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"cpu_load_core" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission",
"nv_util" : "acc_utilization",
"nv_fb_mem_used" : "acc_mem_used",
"nv_power_usage" : "acc_power"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : false,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G",
"cpufreq": "M"
},
"normalize_metrics" : true
}

26
fau-systems/cc-metric-collector/alex/sinks.json Normal file
View File

@@ -0,0 +1,26 @@
{
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "alex",
"password": "XYZ",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"alexstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=alex",
"jwt": "XYZ",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

33
fau-systems/cc-metric-collector/alex/sinks_debug.json Normal file
View File

@@ -0,0 +1,33 @@
{
"fritzganglia": {
"type": "libganglia",
"gmond_config": "/etc/ganglia/gmond.conf",
"libganglia_path": "libganglia.so.0",
"add_ganglia_group": true
},
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "fritz_neu",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"fritzstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=fritz",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s"
},
"debugstdout": {
"type": "stdout",
"output_file": "/tmp/debug.log"
}
}

43
fau-systems/cc-metric-collector/fritz.spr/collectors.json Normal file
View File

@@ -0,0 +1,43 @@
{
"nfs4stat" : {},
"memstat" : {
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"lustrestat" : {
"send_all_metrics" : true,
"use_sudo": false,
"send_diff_values": true,
"send_derived_values": true,
"send_abs_values": false
},
"netstat" : {
"include_devices" : [
"enp1s0",
"enp22s0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"ibstat" : {
"send_abs_values": true,
"send_derived_values": true
},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"cpufreq_cpuinfo": {},
"nfsiostat": {}
}

8
fau-systems/cc-metric-collector/fritz.spr/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/home/hpc/unrz/unrz139/Work/cc-metric-collector/configs/fritz.spr2/sinks2.json",
"collectors" : "/home/hpc/unrz/unrz139/Work/cc-metric-collector/configs/fritz.spr2/collectors.json",
"receivers" : "/home/hpc/unrz/unrz139/Work/cc-metric-collector/configs/fritz.spr2/receivers.json",
"router" : "/home/hpc/unrz/unrz139/Work/cc-metric-collector/configs/fritz.spr2/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/fritz.spr/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

54
fau-systems/cc-metric-collector/fritz.spr/router.json Normal file
View File

@@ -0,0 +1,54 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "fritz",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission",
"cpufreq" : "clock"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : false,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G"
},
"normalize_metrics" : true
}

32
fau-systems/cc-metric-collector/fritz.spr/sinks.json Normal file
View File

@@ -0,0 +1,32 @@
{
"fritzganglia": {
"type": "libganglia",
"gmond_config": "/etc/ganglia/gmond.conf",
"libganglia_path": "libganglia.so.0",
"add_ganglia_group": true
},
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "fritz_neu",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"fritzstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=fritz",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

184
fau-systems/cc-metric-collector/fritz/collectors.json Normal file
View File

@@ -0,0 +1,184 @@
{
"nfs4stat" : {},
"memstat" : {
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"lustrestat" : {
"send_all_metrics" : true,
"use_sudo": false,
"send_diff_values": true,
"send_derived_values": true,
"send_abs_values": false
},
"netstat" : {
"include_devices" : [
"enp1s0",
"enp22s0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"ibstat" : {
"send_abs_values": true,
"send_derived_values": true
},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nfsiostat": {},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"MBOX6C0": "CAS_COUNT_RD",
"MBOX6C1": "CAS_COUNT_WR",
"MBOX7C0": "CAS_COUNT_RD",
"MBOX7C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "0.000001*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0+PMC3*8.0)/time",
"name": "flops_dp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX6C0+MBOX7C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1+MBOX6C1+MBOX7C1)*64.0/time",
"name": "mem_bw",
"unit": "GBytes/s",
"publish": true,
"type": "socket"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"FIXC3": "TOPDOWN_SLOTS",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0+PMC3*16.0)/time",
"name": "flops_sp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
}
]
}
}

8
fau-systems/cc-metric-collector/fritz/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/fritz/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

53
fau-systems/cc-metric-collector/fritz/router.json Normal file
View File

@@ -0,0 +1,53 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "fritz",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : false,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G"
},
"normalize_metrics" : true
}

32
fau-systems/cc-metric-collector/fritz/sinks.json Normal file
View File

@@ -0,0 +1,32 @@
{
"fritzganglia": {
"type": "libganglia",
"gmond_config": "/etc/ganglia/gmond.conf",
"libganglia_path": "libganglia.so.0",
"add_ganglia_group": true
},
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "fritz_neu",
"password": "XYZ",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"fritzstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=fritz",
"jwt": "XYZ",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

33
fau-systems/cc-metric-collector/fritz/sinks_debug.json Normal file
View File

@@ -0,0 +1,33 @@
{
"fritzganglia": {
"type": "libganglia",
"gmond_config": "/etc/ganglia/gmond.conf",
"libganglia_path": "libganglia.so.0",
"add_ganglia_group": true
},
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "fritz_neu",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"fritzstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=fritz",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s"
},
"debugstdout": {
"type": "stdout",
"output_file": "/tmp/debug.log"
}
}

0
fau-systems/cc-metric-collector/meggie-ng/.gitkeep Normal file
View File

161
fau-systems/cc-metric-collector/meggie-ng/collectors.json Normal file
View File

@@ -0,0 +1,161 @@
{
"diskstat" : {},
"ibstat" : {},
"nfs4stat" : {},
"cpustat" : {},
"loadavg" : {},
"iostat" : {},
"netstat" : {
"include_devices" : [
"eno1"
],
"send_abs_values" : true,
"send_derived_values" : true
},
"tempstat" : {
"tag_override" : {
"hwmon1" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon2" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"memstat" : {
"node_stats": true
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0)/time",
"name": "flops_dp",
"publish": true,
"unit": "GFlops/s",
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1)*64.0/time",
"name": "mem_bw",
"publish": true,
"unit": "GBytes/s",
"type": "socket"
},
{
"calc": "PMC0+PMC2",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0)/time",
"name": "flops_sp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"type": "hwthread",
"unit": "%",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"type": "hwthread",
"unit": "GFlops/s",
"publish": true
}
]
}
}

8
fau-systems/cc-metric-collector/meggie-ng/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/meggie-ng/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

53
fau-systems/cc-metric-collector/meggie-ng/router.json Normal file
View File

@@ -0,0 +1,53 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "meggie",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : true,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G"
},
"normalize_metrics" : true
}

26
fau-systems/cc-metric-collector/meggie-ng/sinks.json Normal file
View File

@@ -0,0 +1,26 @@
{
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "meggie",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"fritzstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=meggie",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

167
fau-systems/cc-metric-collector/tinyfat/collectors.bdw.json Normal file
View File

@@ -0,0 +1,167 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0",
"eth1",
"eth2"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon2" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon3" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "pwr_pkg",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "pwr_dram",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0)/time",
"name": "flops_dp",
"publish": true,
"unit": "GFlops/s",
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1)*64.0/time",
"name": "mem_bw",
"publish": true,
"unit": "GBytes/s",
"type": "socket"
},
{
"calc": "PMC0+PMC2",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0)/time",
"name": "flops_sp",
"publish": true,
"unit": "GFlops/s",
"type": "hwthread"
},
{
"calc": "PMC0+PMC2",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"type": "hwthread",
"unit": "GFlops/s",
"publish": true
}
]
}
}

100
fau-systems/cc-metric-collector/tinyfat/collectors.rome.json Normal file
View File

@@ -0,0 +1,100 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0",
"eth1",
"eth2"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon1" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon2" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC1": "ACTUAL_CPU_CLOCK",
"FIXC2": "MAX_CPU_CLOCK",
"PMC0": "RETIRED_INSTRUCTIONS",
"PMC1": "CPU_CLOCKS_UNHALTED",
"PMC2": "RETIRED_SSE_AVX_FLOPS_ALL",
"PMC3": "MERGE",
"DFC0": "DATA_FROM_LOCAL_DRAM_CHANNEL",
"DFC1": "DATA_TO_LOCAL_DRAM_CHANNEL",
"PWR0": "RAPL_CORE_ENERGY",
"PWR1": "RAPL_PKG_ENERGY"
},
"metrics": [
{
"name": "ipc",
"calc": "PMC0/PMC1",
"type": "hwthread",
"publish": true
},
{
"name": "flops_any",
"calc": "1E-9*PMC2/time",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"name": "clock",
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"type": "hwthread",
"unit": "MHz",
"publish": true
},
{
"name": "pwr_core",
"calc": "PWR0/time",
"unit": "Watt",
"type": "socket",
"publish": true
},
{
"name": "pwr_pkg",
"calc": "PWR1/time",
"type": "socket",
"unit": "Watt",
"publish": true
},
{
"name": "mem_bw",
"calc": "1E-9*(DFC0+DFC1)*64.0/time",
"unit": "Gbyte/s",
"type": "socket",
"publish": true
}
]
}
],
"globalmetrics": []
}
}

8
fau-systems/cc-metric-collector/tinyfat/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/tinyfat/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

49
fau-systems/cc-metric-collector/tinyfat/router.json Normal file
View File

@@ -0,0 +1,49 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "tinyfat",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"cpu_load_core" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission",
"pwr_pkg": "cpu_power",
"pwr_dram": "mem_power"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : false,
"num_cache_intervals" : 0
}

26
fau-systems/cc-metric-collector/tinyfat/sinks.json Normal file
View File

@@ -0,0 +1,26 @@
{
"influx": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "tinyfat",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"metricstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=tinyfat",
"jwt": "XYZ",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

30
fau-systems/cc-metric-collector/tinyfat/sinks_debug.json Normal file
View File

@@ -0,0 +1,30 @@
{
"influx": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "tinyfat",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"metricstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=tinyfat",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
},
"debugstdout": {
"type": "stdout",
"output_file": "/tmp/debug.log"
}
}

103
fau-systems/cc-metric-collector/tinygpu/collectors.rome.a100.json Normal file
View File

@@ -0,0 +1,103 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nvidia" : {
"use_pci_info_as_type_id": true,
"process_mig_devices": true
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC1": "ACTUAL_CPU_CLOCK",
"FIXC2": "MAX_CPU_CLOCK",
"PMC0": "RETIRED_INSTRUCTIONS",
"PMC1": "CPU_CLOCKS_UNHALTED",
"PMC2": "RETIRED_SSE_AVX_FLOPS_ALL",
"PMC3": "MERGE",
"DFC0": "DATA_FROM_LOCAL_DRAM_CHANNEL",
"DFC1": "DATA_TO_LOCAL_DRAM_CHANNEL",
"PWR0": "RAPL_CORE_ENERGY",
"PWR1": "RAPL_PKG_ENERGY"
},
"metrics": [
{
"name": "ipc",
"calc": "PMC0/PMC1",
"type": "hwthread",
"publish": true
},
{
"name": "flops_any",
"calc": "1E-9*PMC2/time",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"name": "clock",
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"type": "hwthread",
"unit": "MHz",
"publish": true
},
{
"name": "pwr_core",
"calc": "PWR0/time",
"unit": "Watt",
"type": "socket",
"publish": true
},
{
"name": "pwr_pkg",
"calc": "PWR1/time",
"type": "socket",
"unit": "Watt",
"publish": true
},
{
"name": "mem_bw",
"calc": "1E-9*(DFC0+DFC1)*64.0/time",
"unit": "Gbyte/s",
"type": "socket",
"publish": true
}
]
}
],
"globalmetrics": []
}
}

219
fau-systems/cc-metric-collector/tinygpu/collectors.skx.2080.json Normal file
View File

@@ -0,0 +1,219 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nvidia" : {
"use_pci_info_as_type_id": true,
"process_mig_devices": true
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "0.000001*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0)/time",
"name": "flops_dp1",
"unit": "GFlops/s",
"publish": false,
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1)*64.0/time",
"name": "mem_bw",
"unit": "GBytes/s",
"publish": true,
"type": "socket"
},
{
"calc": "PMC0+PMC2",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0)/time",
"name": "flops_sp1",
"publish": false,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_512B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "PMC0",
"name": "dp_avx_512_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "1E-9*(PMC0*8.0)/time",
"name": "flops_dp2",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC1",
"name": "sp_avx_512_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "1E-9*(PMC1*16.0)/time",
"name": "flops_sp2",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins+dp_avx_512_ins+sp_avx_512_ins)/(sp_fp_ins+dp_fp_ins+dp_avx_512_ins+sp_avx_512_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_sp1+flops_sp2)",
"name": "flops_sp",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp1+flops_dp2)",
"name": "flops_dp",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"calc": "((flops_dp1+flops_dp2) * 2) + (flops_sp1+flops_sp2)",
"name": "flops_any",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
}
]
}
}

175
fau-systems/cc-metric-collector/tinygpu/collectors.skx.3080.json Normal file
View File

@@ -0,0 +1,175 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0",
"eth1",
"eth2"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon1" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon2" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nvidia" : {
"use_pci_info_as_type_id": true,
"process_mig_devices": true
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "0.000001*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0+PMC3*8.0)/time",
"name": "flops_dp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1)*64.0/time",
"name": "mem_bw",
"unit": "GBytes/s",
"publish": true,
"type": "socket"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0+PMC3*16.0)/time",
"name": "flops_sp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
}
]
}
}

219
fau-systems/cc-metric-collector/tinygpu/collectors.skx.v100.json Normal file
View File

@@ -0,0 +1,219 @@
{
"nfs4stat" : {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"cpustat" : {},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"eth0"
],
"send_abs_values": true,
"send_derived_values": true
},
"diskstat" : {},
"iostat" : {},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nvidia" : {
"use_pci_info_as_type_id": true,
"process_mig_devices": true
},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "0.000001*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0)/time",
"name": "flops_dp1",
"unit": "GFlops/s",
"publish": false,
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1)*64.0/time",
"name": "mem_bw",
"unit": "GBytes/s",
"publish": true,
"type": "socket"
},
{
"calc": "PMC0+PMC2",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0)/time",
"name": "flops_sp1",
"publish": false,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_512B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "PMC0",
"name": "dp_avx_512_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "1E-9*(PMC0*8.0)/time",
"name": "flops_dp2",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC1",
"name": "sp_avx_512_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "1E-9*(PMC1*16.0)/time",
"name": "flops_sp2",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins+dp_avx_512_ins+sp_avx_512_ins)/(sp_fp_ins+dp_fp_ins+dp_avx_512_ins+sp_avx_512_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_sp1+flops_sp2)",
"name": "flops_sp",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp1+flops_dp2)",
"name": "flops_dp",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
},
{
"calc": "((flops_dp1+flops_dp2) * 2) + (flops_sp1+flops_sp2)",
"name": "flops_any",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
}
]
}
}

8
fau-systems/cc-metric-collector/tinygpu/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/tinygpu/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

60
fau-systems/cc-metric-collector/tinygpu/router.json Normal file
View File

@@ -0,0 +1,60 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "tinygpu",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"cpu_load_core" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission",
"nv_util" : "acc_utilization",
"nv_fb_mem_used" : "acc_mem_used",
"nv_power_usage" : "acc_power",
"pwr_pkg": "cpu_power",
"pwr_dram": "mem_power"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : false,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G",
"cpufreq": "M"
},
"normalize_metrics" : true
}

26
fau-systems/cc-metric-collector/tinygpu/sinks.json Normal file
View File

@@ -0,0 +1,26 @@
{
"influx": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "tinygpu",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"metricstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=tinygpu",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
}
}

30
fau-systems/cc-metric-collector/tinygpu/sinks_debug.json Normal file
View File

@@ -0,0 +1,30 @@
{
"influx": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "tinygpu",
"password": "XZY",
"ssl": true,
"meta_as_tags": [
"unit"
]
},
"metricstore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=tinygpu",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s",
"flush_delay": "2s",
"max_retries": 1,
"timeout": "10s"
},
"debugstdout": {
"type": "stdout",
"output_file": "/tmp/debug.log"
}
}

176
fau-systems/cc-metric-collector/woody-ng/collectors.icx.json Normal file
View File

@@ -0,0 +1,176 @@
{
"diskstat" : {},
"iostat" : {},
"cpustat": {},
"memstat" : {
"numa_stats": true,
"node_stats": true
},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"enp2s0f0",
"eno1",
"ens2f0"
],
"send_abs_values" : true,
"send_derived_values" : true
},
"tempstat" : {
"tag_override" : {
"hwmon0" : {
"type" : "socket",
"type-id" : "0"
},
"hwmon1" : {
"type" : "socket",
"type-id" : "1"
}
}
},
"nfs4stat" : {},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C0": "CAS_COUNT_RD",
"MBOX0C1": "CAS_COUNT_WR",
"MBOX1C0": "CAS_COUNT_RD",
"MBOX1C1": "CAS_COUNT_WR",
"MBOX2C0": "CAS_COUNT_RD",
"MBOX2C1": "CAS_COUNT_WR",
"MBOX3C0": "CAS_COUNT_RD",
"MBOX3C1": "CAS_COUNT_WR",
"MBOX4C0": "CAS_COUNT_RD",
"MBOX4C1": "CAS_COUNT_WR",
"MBOX5C0": "CAS_COUNT_RD",
"MBOX5C1": "CAS_COUNT_WR",
"MBOX6C0": "CAS_COUNT_RD",
"MBOX6C1": "CAS_COUNT_WR",
"MBOX7C0": "CAS_COUNT_RD",
"MBOX7C1": "CAS_COUNT_WR",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "0.000001*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"publish": true,
"unit": "MHz",
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "cpu_power",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "mem_power",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0+PMC3*8.0)/time",
"name": "flops_dp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C0+MBOX1C0+MBOX2C0+MBOX3C0+MBOX4C0+MBOX5C0+MBOX6C0+MBOX7C0+MBOX0C1+MBOX1C1+MBOX2C1+MBOX3C1+MBOX4C1+MBOX5C1+MBOX6C1+MBOX7C1)*64.0/time",
"name": "mem_bw",
"unit": "GBytes/s",
"publish": true,
"type": "socket"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"FIXC3": "TOPDOWN_SLOTS",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE",
"PMC3": "FP_ARITH_INST_RETIRED_512B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0+PMC3*16.0)/time",
"name": "flops_sp",
"unit": "GFlops/s",
"publish": true,
"type": "hwthread"
},
{
"calc": "PMC0+PMC2+PMC3",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2+PMC3",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"unit": "GFlops/s",
"type": "hwthread",
"publish": true
}
]
}
}

147
fau-systems/cc-metric-collector/woody-ng/collectors.skl.json Normal file
View File

@@ -0,0 +1,147 @@
{
"diskstat" : {},
"iostat" : {},
"cpustat": {},
"memstat" : {
"node_stats": true
},
"loadavg" : {},
"schedstat": {},
"netstat" : {
"include_devices" : [
"enp2s0f0",
"eno1",
"ens2f0"
],
"send_abs_values" : true,
"send_derived_values" : true
},
"tempstat" : {},
"nfs4stat" : {},
"likwid": {
"force_overwrite" : true,
"invalid_to_zero" : true,
"access_mode" : "accessdaemon",
"accessdaemon_path" : "/apps/likwid/system/sbin",
"liblikwid_path": "/apps/likwid/system/lib/liblikwid.so",
"eventsets": [
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"MBOX0C1": "DRAM_READS",
"MBOX0C2": "DRAM_WRITES",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_DOUBLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_DOUBLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_DOUBLE",
"PWR0": "PWR_PKG_ENERGY",
"PWR3": "PWR_DRAM_ENERGY"
},
"metrics": [
{
"calc": "1E-6*(FIXC1/FIXC2)/inverseClock",
"name": "clock",
"unit": "MHz",
"publish": true,
"type": "hwthread"
},
{
"calc": "FIXC0/FIXC1",
"name": "ipc",
"publish": true,
"type": "hwthread"
},
{
"calc": "PWR0/time",
"name": "pwr_pkg",
"unit": "Watt",
"publish": true,
"type": "socket"
},
{
"calc": "PWR3/time",
"name": "pwr_dram",
"publish": true,
"unit": "Watt",
"type": "socket"
},
{
"calc": "1E-9*(PMC0*2.0+PMC1+PMC2*4.0)/time",
"name": "flops_dp",
"publish": true,
"unit": "GFlops/s",
"type": "hwthread"
},
{
"calc": "1E-9*(MBOX0C1+MBOX0C2)*64.0/time",
"name": "mem_bw",
"publish": true,
"unit": "GBytes/s",
"type": "socket"
},
{
"calc": "PMC0+PMC2",
"name": "dp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "dp_fp_ins",
"type": "hwthread",
"publish": false
}
]
},
{
"events": {
"FIXC0": "INSTR_RETIRED_ANY",
"FIXC1": "CPU_CLK_UNHALTED_CORE",
"FIXC2": "CPU_CLK_UNHALTED_REF",
"PMC0": "FP_ARITH_INST_RETIRED_128B_PACKED_SINGLE",
"PMC1": "FP_ARITH_INST_RETIRED_SCALAR_SINGLE",
"PMC2": "FP_ARITH_INST_RETIRED_256B_PACKED_SINGLE"
},
"metrics": [
{
"calc": "1E-9*(PMC0*4.0+PMC1+PMC2*8.0)/time",
"name": "flops_sp",
"publish": true,
"unit": "GFlops/s",
"type": "hwthread"
},
{
"calc": "PMC0+PMC2",
"name": "sp_vec_ins",
"type": "hwthread",
"publish": false
},
{
"calc": "PMC0+PMC1+PMC2",
"name": "sp_fp_ins",
"type": "hwthread",
"publish": false
}
]
}
],
"globalmetrics": [
{
"calc": "100*((sp_vec_ins+dp_vec_ins)/(sp_fp_ins+dp_fp_ins))",
"name": "vectorization_ratio",
"unit": "%",
"type": "hwthread",
"publish": true
},
{
"calc": "(flops_dp * 2) + flops_sp",
"name": "flops_any",
"type": "hwthread",
"unit": "GFlops/s",
"publish": true
}
]
}
}

8
fau-systems/cc-metric-collector/woody-ng/config.json Normal file
View File

@@ -0,0 +1,8 @@
{
"sinks": "/etc/cc-metric-collector/sinks.json",
"collectors" : "/etc/cc-metric-collector/collectors.json",
"receivers" : "/etc/cc-metric-collector/receivers.json",
"router" : "/etc/cc-metric-collector/router.json",
"interval": "60s",
"duration": "10s"
}

1
fau-systems/cc-metric-collector/woody-ng/receivers.json Normal file
View File

@@ -0,0 +1 @@
{}

54
fau-systems/cc-metric-collector/woody-ng/router.json Normal file
View File

@@ -0,0 +1,54 @@
{
"add_tags" : [
{
"key" : "cluster",
"value" : "woodyng",
"if" : "*"
}
],
"rename_metrics" : {
"load_one" : "cpu_load",
"cpu_load_core" : "cpu_load",
"net_bytes_in_bw" : "net_bytes_in",
"net_bytes_out_bw" : "net_bytes_out",
"net_pkts_in_bw" : "net_pkts_in",
"net_pkts_out_bw" : "net_pkts_out",
"ib_recv_bw" : "ib_recv",
"ib_xmit_bw" : "ib_xmit",
"ib_recv_pkts_bw": "ib_recv_pkts",
"ib_xmit_pkts_bw": "ib_xmit_pkts",
"lustre_read_bytes_diff" : "lustre_read_bytes",
"lustre_read_requests_diff" : "lustre_read_requests",
"lustre_write_bytes_diff" : "lustre_write_bytes",
"lustre_write_requests_diff" : "lustre_write_requests",
"lustre_open_diff" : "lustre_open",
"lustre_close_diff" : "lustre_close",
"lustre_setattr_diff" : "lustre_setattr",
"lustre_getattr_diff" : "lustre_getattr",
"lustre_statfs_diff": "lustre_statfs",
"lustre_inode_permission_diff" : "lustre_inode_permission"
},
"drop_metrics" : [
"net_bytes_in",
"net_bytes_out",
"ib_recv",
"ib_xmit",
"ib_recv_pkts",
"ib_xmit_pkts",
"net_pkts_in",
"net_pkts_out",
"lustre_read_bytes",
"lustre_read_requests",
"lustre_write_bytes",
"lustre_write_requests"
],
"interval_timestamp" : true,
"num_cache_intervals" : 0,
"change_unit_prefix": {
"mem_used": "G",
"swap_used": "G",
"mem_total": "G",
"swap_total": "G"
},
"normalize_metrics" : true
}

20
fau-systems/cc-metric-collector/woody-ng/sinks.json Normal file
View File

@@ -0,0 +1,20 @@
{
"nhrinflux": {
"type": "influxasync",
"host": "monitoring-test.nhr.uni-erlangen.de",
"port": "8086",
"organization": "ClusterCockpit",
"database": "woodyng",
"password": "XZY",
"ssl": true
},
"woodystore": {
"type": "http",
"url": "http://monitoring.nhr.fau.de:8082/api/write?cluster=woodyng",
"jwt": "XZY",
"meta_as_tags": [
"unit"
],
"idle_connection_timeout": "60s"
}
}