mirror of
https://github.com/ClusterCockpit/cc-specifications.git
synced 2024-11-10 01:37:25 +01:00
490 lines
15 KiB
JSON
490 lines
15 KiB
JSON
{
|
|
"$schema": "http://json-schema.org/draft/2020-12/schema",
|
|
"$id": "job-data.schema.json",
|
|
"title": "Job metric data list",
|
|
"description": "Collection of metric data of a HPC job",
|
|
"type": "object",
|
|
"properties": {
|
|
"mem_used": {
|
|
"description": "Memory capacity used",
|
|
"type": "object",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"flops_any": {
|
|
"description": "Total flop rate with DP flops scaled up",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"mem_bw": {
|
|
"description": "Main memory bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"net_bw": {
|
|
"description": "Total fast interconnect network bandwidth",
|
|
"type": "object",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"ipc": {
|
|
"description": "Instructions executed per cycle",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"cpu_used": {
|
|
"description": "CPU active core utilization",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"cpu_load": {
|
|
"description": "CPU requested core utilization (load 1m)",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"flops_dp": {
|
|
"description": "Double precision flop rate",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"flops_sp": {
|
|
"description": "Single precision flops rate",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"vectorization_ratio": {
|
|
"description": "Fraction of arithmetic instructions using SIMD instructions",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"cpu_power": {
|
|
"description": "CPU power consumption",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"mem_power": {
|
|
"description": "Memory power consumption",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"acc_utilization": {
|
|
"description": "GPU utilization",
|
|
"properties": {
|
|
"accelerator": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"accelerator"
|
|
]
|
|
},
|
|
"acc_mem_used": {
|
|
"description": "GPU memory capacity used",
|
|
"properties": {
|
|
"accelerator": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"accelerator"
|
|
]
|
|
},
|
|
"acc_power": {
|
|
"description": "GPU power consumption",
|
|
"properties": {
|
|
"accelerator": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"accelerator"
|
|
]
|
|
},
|
|
"clock": {
|
|
"description": "Average core frequency",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"socket": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"memoryDomain": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"core": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
},
|
|
"hwthread": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"minProperties": 1
|
|
},
|
|
"eth_read_bw": {
|
|
"description": "Ethernet read bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"eth_write_bw": {
|
|
"description": "Ethernet write bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"filesystems": {
|
|
"description": "Array of filesystems",
|
|
"type": "array",
|
|
"items": {
|
|
"type": "object",
|
|
"properties": {
|
|
"name": {
|
|
"type": "string"
|
|
},
|
|
"type": {
|
|
"type": "string",
|
|
"enum": [
|
|
"nfs",
|
|
"lustre",
|
|
"gpfs",
|
|
"nvme",
|
|
"ssd",
|
|
"hdd",
|
|
"beegfs"
|
|
]
|
|
},
|
|
"read_bw": {
|
|
"description": "File system read bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"write_bw": {
|
|
"description": "File system write bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"read_req": {
|
|
"description": "File system read requests",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"write_req": {
|
|
"description": "File system write requests",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"inodes": {
|
|
"description": "File system write requests",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"accesses": {
|
|
"description": "File system open and close",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"fsync": {
|
|
"description": "File system fsync",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"create": {
|
|
"description": "File system create",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"open": {
|
|
"description": "File system open",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"close": {
|
|
"description": "File system close",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"seek": {
|
|
"description": "File system seek",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
}
|
|
},
|
|
"required": [
|
|
"name",
|
|
"type",
|
|
"read_bw",
|
|
"write_bw"
|
|
]
|
|
},
|
|
"minItems": 1
|
|
}
|
|
},
|
|
"ic_rcv_packets": {
|
|
"description": "Network interconnect read packets",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"ic_send_packets": {
|
|
"description": "Network interconnect send packet",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"ic_read_bw": {
|
|
"description": "Network interconnect read bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"ic_write_bw": {
|
|
"description": "Network interconnect write bandwidth",
|
|
"properties": {
|
|
"node": {
|
|
"$ref": "job-metric-data.schema.json"
|
|
}
|
|
},
|
|
"required": [
|
|
"node"
|
|
]
|
|
},
|
|
"required": [
|
|
"cpu_used",
|
|
"mem_used",
|
|
"flops_any",
|
|
"mem_bw",
|
|
"net_bw",
|
|
"filesystems"
|
|
]
|
|
}
|