(Optional) Configuring Monitoring and Alarms

Monitoring Configurations

Using Cloud Eye to Monitor NPU Resources

Lite Server uses Cloud Eye for monitoring. Its required agent plug-in comes pre-installed. You can check NPU resource data directly through Cloud Eye. For details, see Monitoring Lite Server Resources.

Key Metrics

The table below lists the common monitoring metrics. For details about all supported metrics, see Lite Server Monitoring Metrics.

**Table 1** Common monitoring metrics
No.	Category	Metric	Display Name	Description	Unit	Number System	Range
1	DDR	npu_util_rate_mem	NPU Memory Usage	NPU memory usage	%	N/A	0%–100%
2	DDR	npu_util_rate_mem_bandwidth	NPU Memory Bandwidth Usage	NPU memory bandwidth usage	%	N/A	0%–100%
3	HBM	npu_hbm_bandwidth_util	HBM Bandwidth Usage	NPU HBM bandwidth usage	%	N/A	0%–100%
4	HBM	npu_util_rate_hbm_bw	HBM Bandwidth Usage	NPU HBM bandwidth usage	%	N/A	0%–100%
5	AI Core	npu_util_rate_ai_core	NPU AI Core Usage	AI core usage of the NPU	%	N/A	0%–100%
6	/	mem_usedPercent	Memory Usage	Memory usage of the monitored object. Linux: Obtain the metric value from the /proc/meminfo file: (MemTotal - MemAvailable)/MemTotal If MemAvailable is displayed in /proc/meminfo, MemUsedPercent = (MemTotal-MemAvailable)/MemTotal If MemAvailable is not displayed in /proc/meminfo, MemUsedPercent = (MemTotal – MemFree – Buffers – Cached)/MemTotal Collection method (Windows): formula (Used memory size/Total memory size x 100%)	%	N/A	0%–100%

Alarm Settings

You can use Cloud Eye to collect key events and cloud resource operation events. When an event occurs, you will receive an alarm. You can check supported events in Supported Events.

Parent topic: Adapting Mainstream Open-Source Models to AscendFactory NPU Training Based on Lite Server

Previous topic: Collecting and Storing Logs

Next topic: Configuration Optimization and Fault Recovery