PA HCCS Bandwidth Monitoring

The config_pa.yaml configuration file in the System Methodology Profiler tool package can be used to collect PA events, continuously monitoring the HCCS bandwidth on a server.

1. View PA events.

   cd /path_to_ksys
   ./ksys collect -i 5 -c config_pa.yaml -o /data_path

   

   • /path_to_ksys: specifies the System Methodology Profiler tool installation path.
   • -o: specifies the path to the generated pa_data.json file. This file can be connected to the open source Prometheus dashboard.
   • -i: refreshes HCCS monitoring data at a collection interval of 5 seconds.
   • -c: specifies the config_pa.yaml configuration file.
2. View the pa_data.json file.
   The format of each time series data point in pa_data.json is as follows:

   {
   "version": "1.0",
   "name": "hccs_bandwidth",
   "desc": "hccs bandwidth info, unit is 'MB/s'.",
     "timestamp": 1748484451170,
     "data_list": [
   {
   "value": 500.75,
   "label": {
   "numa": "0",
   "device": "hisi_sicl2_pa0",
   "link": "1",
   "direction": "out", 
   "path": "P0->P1"
   }
       },
   ......
     ]
   }

   • version: specifies the JSON file version that supports the HCCS monitoring capability.
   • name: specifies the metric name. hccs_bandwidth indicates that the obtained indicator is HCCS bandwidth.
   • desc: specifies further description of the metric.
   • timestamp: records the time when the current file data is updated.
   • data_list: specific data of each device. The value in each element is the HCCS bandwidth of a device with the minimum granularity. The unit is MB/s in desc. label is an extensible k-v dictionary. For details, see Table 1.

     Table 1 Fields in label

     Parameter	Description
     numa	NUMA node to which the current data belongs. A NUMA node can have multiple devices.
     device	PA device to which the current data belongs. A device can have multiple links.
     link	ID of the link to which the current data belongs. link is the minimum device granularity for data collection.
     direction	Current data flow direction, which is out or in. out: Data flows out of the current CPU socket to another CPU socket. in: Data flows from another CPU socket to the current CPU socket.
     path	Path of the current data. For example, P0->P1 indicates that the current socket is CPU socket 0, which from data flow to CPU socket 1.