What Metrics Are Supported by the Agent?

cpu_usage

(Agent) CPU Usage

Used to monitor CPU usage.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) value.
• Collection method (Windows): Obtain the metric value using the API GetSystemTimes.

0–100

%

N/A

2.4.1

1 minute

cpu_usage_idle

(Agent) Idle CPU Usage

Percentage of the time that CPU is idle.

Unit: Percent

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period.
• Collection method (Windows): Obtain the metric value using the API GetSystemTimes.

0–100

%

N/A

2.4.5

1 minute

cpu_usage_other

(Agent) Other Process CPU Usage

Other CPU usage of the monitored object.

• Collection method (Linux): Other Process CPU Usage = 1– Idle CPU Usage – Kernel Space CPU Usage – User Space CPU Usage
• Collection method (Windows): Other Process CPU Usage = 1– Idle CPU Usage – Kernel Space CPU Usage – User Space CPU Usage

0–100

%

N/A

2.4.5

1 minute

cpu_usage_system

(Agent) Kernel Space CPU Usage

Percentage of time that the CPU is used by kernel space.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) sy value.
• Collection method (Windows): Obtain the metric value using the API GetSystemTimes.

0–100

%

N/A

2.4.5

1 minute

cpu_usage_user

(Agent) User Space CPU Usage

Percentage of time that the CPU is used by user space.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) us value.
• Collection method (Windows): Obtain the metric value using the API GetSystemTimes.

0–100

%

N/A

2.4.5

1 minute

cpu_usage_nice

(Agent) Nice Process CPU Usage

CPU usage of the nice process of the measured object.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) ni value.
• Collection method (Windows): not supported

0–100

%

N/A

2.4.5

1 minute

cpu_usage_iowait

(Agent) iowait Process CPU Usage

CPU usage of the measured object in the iowait state.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) wa value.
• Collection method (Windows): not supported

0–100

%

N/A

2.4.5

1 minute

cpu_usage_irq

(Agent) CPU Interrupt Time

Percentage of time that the CPU is servicing interrupts.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) hi value.
• Collection method (Windows): not supported

0–100

%

N/A

2.4.5

1 minute

cpu_usage_softirq

(Agent) CPU Software Interrupt Time

Percentage of time that the CPU is servicing software interrupts.

• Collection method (Linux): Check the metric value changes in file /proc/stat in a collection period. You can run the top command to check the %Cpu(s) si value.
• Collection method (Windows): not supported

0–100

%

N/A

2.4.5

1 minute

load_total_average1

(Agent) 1-Minute Load Average

CPU load averaged from the last one minute.

• Collection method (Linux): The value of this metric is that of load1 in /proc/loadavg. You can run the top command to check the value of load1.
• Collection method (Windows): not supported

≥ 0

None

2.8.2

1 minute

load_total_average5

(Agent) 5-Minute Load Average

CPU load averaged from the last five minutes.

• Collection method (Linux): The value of this metric is that of load5 in /proc/loadavg. You can run the top command to check the value of load5.
• Collection method (Windows): not supported

≥ 0

None

2.8.2

1 minute

load_total_average15

(Agent) 15-Minute Load Average

CPU load averaged from the last 15 minutes.

• Collection method (Linux): The value of this metric is that of load15 in /proc/loadavg. You can run the top command to check the value of load15.
• Collection method (Windows): not supported

≥ 0

None

2.8.2

1 minute

load_average1

(Agent) 1-Minute Single-Core Load Average

Single-core CPU load averaged from the last one minute.

• Collection method (Linux): The value of this metric is the result of the value of load1 in /proc/loadavg divided by the number of logical CPUs. You can run the top command to check the value of load1.
• Collection method (Windows): not supported

≥ 0

None

2.4.1

1 minute

load_average5

(Agent) 5-Minute Single-Core Load Average

Single-core CPU load averaged from the last five minutes.

• Collection method (Linux): The value of this metric is the result of the value of load5 in /proc/loadavg divided by the number of logical CPUs. You can run the top command to check the value of load5.
• Collection method (Windows): not supported

≥ 0

None

2.4.1

1 minute

load_average15

(Agent) 15-Minute Single-Core Load Average

Single-core CPU load averaged from the last 15 minutes.

• Collection method (Linux): The value of this metric is the result of the value of load15 in /proc/loadavg divided by the number of logical CPUs. You can run the top command to check the value of load15.
• Collection method (Windows): not supported

≥ 0

None

2.4.1

1 minute

mem_available

(Agent) Available Memory

Amount of memory that is available and can be given instantly to processes.

• Collection method (Linux): Obtain the metric value from /proc/meminfo.

  If MemAvailable is displayed in /proc/meminfo, obtain the value.

  If MemAvailable is not displayed in /proc/meminfo, MemAvailable = MemFree + Buffers + Cached

• Collection method (Windows): formula (Total memory – Used memory) The value is obtained by calling the Windows API GlobalMemoryStatusEx.

≥ 0

GB

N/A

2.4.5

1 minute

mem_usedPercent

(Agent) Memory Usage

Memory usage of the instance.

• Collection method (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%)

0–100

%

N/A

2.4.1

1 minute

mem_free

(Agent) Idle Memory

Amount of memory that is not being used.

• Collection method (Linux): Obtain the metric value from /proc/meminfo.
• Collection method (Windows): not supported

≥ 0

GB

N/A

2.4.5

1 minute

mem_buffers

(Agent) Buffer

Amount of memory that is being used for buffers.

• Collection method (Linux): Obtain the metric value from /proc/meminfo. You can run the top command to check the KiB Mem:buffers value.
• Collection method (Windows): not supported

≥ 0

GB

N/A

2.4.5

1 minute

mem_cached

(Agent) Cache

Amount of memory that is being used for file caches.

• Collection method (Linux): Obtain the metric value from /proc/meminfo. You can run the top command to check the KiB Swap:cached Mem value.
• Collection method (Windows): not supported

≥ 0

GB

N/A

2.4.5

1 minute

total_open_files

(Agent) Total File Handles

Total handles used by all processes.

• Collection method (Linux): Use the /proc/{pid}/fd file to summarize the handles used by all processes.
• Collection method (Windows): not supported

≥ 0

Count

N/A

2.4.5

1 minute

disk_free

(Agent) Available Disk Space

Free space on the disks.

• Collection method (Linux): Run the df -h command to check the value in the Avail column. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): Use the Windows Management Instrumentation (WMI) API GetDiskFreeSpaceExW to obtain disk space data. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

≥ 0

GB

N/A

2.4.1

1 minute

disk_total

(Agent) Disk Storage Capacity

Total disk capacity.

• Collection method (Linux): Run the df -h command to check the value in the Size column.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): Use the WMI API GetDiskFreeSpaceExW to obtain disk space data. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

≥ 0

GB

N/A

2.4.5

1 minute

disk_used

(Agent) Used Disk Space

Disk's used space.

• Collection method (Linux): Run the df -h command to check the value in the Used column. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): Use the WMI API GetDiskFreeSpaceExW to obtain disk space data. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

≥ 0

GB

N/A

2.4.5

1 minute

disk_usedPercent

(Agent) Disk Usage

Percentage of used disk space. It is calculated as follows: Disk Usage = Used Disk Space/Disk Storage Capacity.

• Collection method (Linux): It is calculated as follows: Used/Size. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): Use the WMI API GetDiskFreeSpaceExW to obtain disk space data. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

0–100

%

N/A

2.4.1

1 minute

disk_rwstate

(Agent) Disk Read/Write Status

Read and write status of the disk attached to the monitored object. The status can be 0 (read and write) or 1 (read-only).

• Collection method (Linux): Obtain the disk attachment status from the /proc/1/mountinfo file.
• Collection method (Windows): not supported

• 0: read and write
• 1: read-only

None

N/A

2.5.6

1 minute

disk_agt_read_bytes_rate

(Agent) Disks Read Rate

Volume of data read from the instance per second.

• Collection method (Linux):

  Calculate the data changes in the sixth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows):
  • The disk I/O data is obtained through the Win32_PerfFormattedData_PerfDisk_LogicalDisk object in WMI. The object is obtained once in each collection period. The instantaneous value returned by the object indicates the metric value in a collection period.
  • The mount point prefix cannot exceed 64 characters. It must start with a letter and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
  • When the CPU usage is high, monitoring data obtaining timeout may occur and monitoring data cannot be obtained.

≥ 0

byte/s

1024(IEC)

2.4.5

1 minute

disk_agt_read_requests_rate

(Agent) Disks Read Requests

Number of read requests sent to the monitored disk per second.

• Collection method (Linux):

  The disk read requests are calculated by measuring the data changes in the fourth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows):
  • The disk I/O data is obtained through the Win32_PerfFormattedData_PerfDisk_LogicalDisk object in WMI. The object is obtained once in each collection period. The instantaneous value returned by the object indicates the metric value in a collection period.
  • The mount point prefix cannot exceed 64 characters. It must start with a letter and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
  • When the CPU usage is high, monitoring data obtaining timeout may occur and monitoring data cannot be obtained.

≥ 0

Request/s

N/A

2.4.5

1 minute

disk_agt_write_bytes_rate

(Agent) Disks Write Rate

Volume of data written to the instance per second.

• Collection method (Linux):

  The disk write rate is calculated by measuring the data changes in the tenth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows):
  • The disk I/O data is obtained through the Win32_PerfFormattedData_PerfDisk_LogicalDisk object in WMI. The object is obtained once in each collection period. The instantaneous value returned by the object indicates the metric value in a collection period.
  • The mount point prefix cannot exceed 64 characters. It must start with a letter and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
  • When the CPU usage is high, monitoring data obtaining timeout may occur and monitoring data cannot be obtained.

≥ 0

byte/s

1024(IEC)

2.4.5

1 minute

disk_agt_write_requests_rate

(Agent) Disks Write Requests

Number of write requests sent to the monitored disk per second.

• Collection method (Linux):

  The disk write rate is calculated by measuring the data changes in the eighth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows):
  • The disk I/O data is obtained through the Win32_PerfFormattedData_PerfDisk_LogicalDisk object in WMI. The object is obtained once in each collection period. The instantaneous value returned by the object indicates the metric value in a collection period.
  • The mount point prefix cannot exceed 64 characters. It must start with a letter and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
  • When the CPU usage is high, monitoring data obtaining timeout may occur and monitoring data cannot be obtained.

≥ 0

Request/s

N/A

2.4.5

1 minute

disk_readTime

(Agent) Average Read Request Time

The average time taken for disk read operations.

• Collection method (Linux):

  The average read request time is calculated by measuring the data changes in the seventh column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

ms/count

N/A

2.4.5

1 minute

disk_writeTime

(Agent) Average Write Request Time

The average time taken for disk write operations.

• Collection method (Linux):

  The average write request time is calculated by measuring the data changes in the eleventh column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

ms/count

N/A

2.4.5

1 minute

disk_ioUtils

(Agent) Disk I/O Usage

Disk I/O usage of the measured object.

• Collection method (Linux):

  The disk I/O usage is calculated by measuring the data changes in the thirteenth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

0–100

%

N/A

2.4.1

1 minute

disk_queue_length

(Agent) Disk Queue Length

Average number of read or write requests queued up for completion for the monitored disk in the monitoring period.

• Collection method (Linux):

  The average disk queue length is calculated by measuring the data changes in the fourteenth column of the corresponding device in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows):
  • The disk I/O data is obtained through the Win32_PerfFormattedData_PerfDisk_LogicalDisk object in WMI. The object is obtained once in each collection period. The instantaneous value returned by the object indicates the metric value in a collection period.
  • The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
  • When the CPU usage is high, monitoring data obtaining timeout may occur and monitoring data cannot be obtained.

≥ 0

count

N/A

2.4.5

1 minute

disk_write_bytes_per_operation

(Agent) Disk Bytes Per Write Operation

Average number of bytes in an I/O write for the monitored disk in the monitoring period.

• Collection method (Linux):

  The average disk write size is calculated by measuring the data changes in the tenth column of the corresponding device to divide that of the eighth column in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Byte/op

N/A

2.4.5

1 minute

disk_read_bytes_per_operation

(Agent) Disk Bytes Per Read Operation

Average number of bytes in an I/O read for the monitored disk in the monitoring period.

• Collection method (Linux):

  The average disk read size is calculated by using the data changes in the sixth column of the corresponding device to divide that of the fourth column in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Byte/op

N/A

2.4.5

1 minute

disk_io_svctm

(Agent) Disk I/O Svctm

Average time in an I/O read or write for the monitored disk in the monitoring period.

• Collection method (Linux):

  The average disk I/O service time is calculated by using the data changes in the thirteenth column of the corresponding device to divide the sum of data changes in the fourth and eighth columns in file /proc/diskstats in a collection period.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

ms/op

N/A

2.4.5

1 minute

disk_device_used_percent

(Agent) Block Device Usage

Percentage of total disk space that is used. The calculation formula is as follows: Used storage space of all mounted disk partitions/Total disk storage space.

• Collection method (Linux): Summarize the disk usage of each mount point, calculate the total disk size based on the disk sector size and number of sectors, and calculate the overall disk usage.
• Currently, Windows does not support this metric.
  NOTE:

  This metric is disabled by default. For details about how to enable it, see How Do I Enable or Disable Metric Collection by Modifying the Configuration File?

0–100

%

N/A

2.5.6

1 minute

disk_agt_max_read_bytes_rate

(Agent) Max Disk Read Speed

The maximum number of bytes read from the monitored object per second in a collection period.

• Collection method (Linux): The system calculates the change of the sixth column (sectors read) in the /proc/diskstats file per second to obtain the disk read speed. The maximum value in each period is used as the maximum disk read speed.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Byte/s

1024(IEC)

2.8.10

1 minute

disk_agt_max_read_requests_rate

(Agent) Max Disk Read Request Speed

The maximum number of read requests to the monitored object per second in a collection period.

• Collection method (Linux): The disk read rate is calculated based on the change of the fourth column (IOPS read) in the /proc/diskstats file per second. The maximum value in each period is used as the maximum disk read rate.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Request/s

N/A

2.8.10

1 minute

disk_agt_max_write_bytes_rate

(Agent) Max Disk Write Speed

The maximum number of bytes written to the monitored object per second in a collection period.

• Collection method (Linux): The disk write rate is calculated based on the change of the tenth column (sectors write) in the /proc/diskstats file per second. The maximum value in each period is used as the maximum disk write rate.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Byte/s

1024(IEC)

2.8.10

1 minute

disk_agt_max_write_requests_rate

(Agent) Max Disk Write Request Speed

The maximum number of write requests to the monitored object per second in a collection period.

• Collection method (Linux): The disk write rate is calculated based on the change of the eighth column (IOPS write) in the /proc/diskstats file per second. The maximum value in each period is used as the maximum disk write rate.

  The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).

• Collection method (Windows): not supported

≥ 0

Request/s

N/A

2.8.10

1 minute

disk_fs_rwstate

(Agent) File System Read/Write Status

Read and write status of the mounted file system of the monitored object. The status can be 0 (read and write) or 1 (read-only).

• Collection method (Linux): Check file system information in the fourth column in file /proc/mounts.
• Collection method (Windows): not supported

• 0: read and write
• 1: read-only

None

N/A

2.4.5

1 minute

disk_inodesTotal

(Agent) Disk inode Total

Total number of index nodes on the disk.

• Collection method (Linux): Run the df -i command to check the value in the Inodes column. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.5

1 minute

disk_inodesUsed

(Agent) Total inode Used

Number of used index nodes on the disk.

• Collection method (Linux): Run the df -i command to check the value in the IUsed column. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.5

1 minute

disk_inodesUsedPercent

(Agent) Percentage of Total inode Used

Number of used index nodes on the disk

• Collection method (Linux): Run the df -i command to check the value in the IUse% column. The path of the mount point prefix cannot exceed 64 characters. It must start with a letter, and contain only digits, letters, hyphens (-), periods (.), and swung dashes (~).
• Collection method (Windows): not supported

0–100

%

N/A

2.4.1

1 minute

ntp_offset

(Agent) NTP Offset

NTP offset of the monitored object.

• Collection method (Linux): Run the ntpq -p or chronyc sources -v command.
• Collection method (Windows): not supported

≥ 0

ms

2.7.1

1 minute

net_tcp_total

(Agent) Total TCP Connections

Total number of TCP connections.

• Collection method (Linux): Run the ss -s command to view the total number of TCP connections.
  NOTE:

  The number of TCP connections refers to the number of all active TCP connections on an ECS.

• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.1

1 minute

net_tcp_established

(Agent) TCP ESTABLISHED Connections

Number of TCP connections in the ESTABLISHED state.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.1

1 minute

net_udp_total

(Agent) Total UDP Connections

≥ 0

count

N/A

2.6.4

1 minute

net_tcp_sys_sent

(Agent) TCP SYS_SENT Connections

Number of TCP connections that are being requested by the client.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_sys_recv

(Agent) TCP SYS_RECV Connections

Number of pending TCP connections received by the server.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_fin_wait1

(Agent) TCP FIN_WAIT1 Connections

Number of TCP connections waiting for ACK packets when the connections are being actively closed by the client

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_fin_wait2

(Agent) TCP FIN_WAIT2 Connections

Number of TCP connections in the FIN_WAIT2 state.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_time_wait

(Agent) TCP TIME_WAIT Connections

Number of TCP connections in the TIME_WAIT state.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_close

(Agent) TCP CLOSE Connections

Number of closed TCP connections.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_close_wait

(Agent) TCP CLOSE_WAIT Connections

Number of TCP connections in the CLOSE_WAIT state.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_last_ack

(Agent) TCP LAST_ACK Connections

Number of TCP connections waiting for ACK packets when the connections are being passively closed by the client.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_listen

(Agent) TCP LISTEN Connections

Number of TCP connections in the LISTEN state.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_closing

(Agent) TCP CLOSING Connections

Number of TCP connections to be automatically closed by the server and the client at the same time.

• Collection method (Linux): Obtain TCP connections in all states from the /proc/net/tcp file, and then collect the number of connections in each state.
• Collection method (Windows): Obtain the metric value using the GetTcpTable2 API.

≥ 0

count

N/A

2.4.5

1 minute

net_tcp_retrans

(Agent) TCP Retransmission Rate

Percentage of packets that are resent.

• Collection method (Linux): Obtain the metric value from the /proc/net/snmp file. The value is the ratio of the number of sent packets to the number of retransmitted packages in a collection period.
• Collection method (Windows): Obtain the metric value using the GetTcpStatistics API.

0–100

%

N/A

2.4.5

1 minute

net_bitRecv

(Agent) Outbound Bandwidth

Number of bits sent by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data.

≥ 0

bit/s

1024(IEC)

2.4.1

1 minute

net_bitSent

(Agent) Inbound Bandwidth

Number of bits received by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data.

≥ 0

bit/s

1024(IEC)

2.4.1

1 minute

net_packetRecv

(Agent) NIC Packet Receive Rate

Number of packets received by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data.

≥ 0

Count/s

N/A

2.4.1

1 minute

net_packetSent

(Agent) NIC Packet Send Rate

Number of packets sent by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data.

≥ 0

Count/s

N/A

2.4.1

1 minute

net_errin

(Agent) Receive Error Rate

Percentage of receive errors detected by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data. You can run the netstat -e command to view the number of error packets.

0–100

%

N/A

2.4.5

1 minute

net_errout

(Agent) Transmit Error Rate

Percentage of transmit errors detected by this NIC per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data. You can run the netstat -e command to view the number of error packets.

0–100

%

N/A

2.4.5

1 minute

net_dropin

(Agent) Received Packet Drop Rate

Percentage of packets received by this NIC which were dropped per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data. You can run the netstat -e command to view the number of dropped packets.

0–100

%

N/A

2.4.5

1 minute

net_dropout

(Agent) Transmitted Packet Drop Rate

Percentage of packets transmitted by this NIC which were dropped per second.

• Collection method (Linux): Check metric value changes in file /proc/net/dev in a collection period.
• Collection method (Windows): Use the MibIfRow object in WMI to obtain network metric data. You can run the netstat -e command to view the number of dropped packets.

0–100

%

N/A

2.4.5

1 minute

block_device_used_percent

(Agent) Block Storage Usage

Block device usage of the monitored object.

• Collection method (Linux): Calculate the sum of storage space used by all mount points. Calculate the total disk storage space based on the disk sector size and the number of sectors. Then, calculate the block device storage usage.
• Collection method (Windows): Call the Windows API GetDiskFreeSpaceEx to obtain the used space of each path. Use WMI to query Win32_DiskDrive to obtain the total capacity of each physical disk. Then, calculate the block device storage usage.

0–100

%

2.8.6

1 minute

block_device_used

(Agent) Used Block Storage Capacity

Used block device storage capacity of the monitored object.

• Collection method (Linux): Sum up the storage space used by all mount points.
• Collection method (Windows): Call the Windows API GetDiskFreeSpaceEx to obtain the space used by each physical disk path.

≥ 0

GiB

2.8.6

1 minute

block_device_total

(Agent) Total Block Storage Capacity

Total block device storage capacity of the monitored object.

• Collection method (Linux): Calculate the total disk storage space based on the disk sector size and the number of sectors.
• Collection method (Windows): Use WMI to query Win32_DiskDrive to obtain the total size of each physical disk.

≥ 0

GiB

2.8.6

1 minute

proc_running_count

(Agent) Running Processes

Number of processes that are running.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.1

1 minute

proc_idle_count

(Agent) Idle Processes

Number of processes that are idle.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.1

1 minute

proc_zombie_count

(Agent) Zombie Processes

Number of zombie processes.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.1

1 minute

proc_blocked_count

(Agent) Blocked Processes

Number of processes that are blocked.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.1

1 minute

proc_sleeping_count

(Agent) Sleeping Processes

Number of processes that are sleeping.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): not supported

≥ 0

None

N/A

2.4.1

1 minute

proc_total_count

(Agent) Total Processes

Total number of processes on the monitored object.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): Obtain the total number of processes by using the system process status support module psapi.dll.

≥ 0

None

N/A

2.4.1

1 minute

proc_specified_count

(Agent) Specified Processes

Number of specified processes.

• Collection method (Linux): You can obtain the state of each process by checking the Status value in the /proc/pid/status file, and then collect the total number of processes in each state.
• Collection method (Windows): Obtain the total number of processes by using the system process status support module psapi.dll.

≥ 0

None

N/A

2.4.1

1 minute

specified_process_cpu

(Agent)Process CPU Usage

CPU usage of a specified process of the monitored object.

• Collection method (Linux): Check the metric value changes in file /proc/pid/stat.
• Collection method (Windows): Call the Windows API GetProcessTimes to obtain the CPU usage of the process.

0 to 100 x Number of vCPUs

%

N/A

2.7.2

1 minute

specified_process_mem

(Agent)Process Memory Usage

Memory usage of a specified process of the monitored object.

• Collection method (Linux):

  RSS x PAGESIZE/MemTotal

  RSS: obtained from the second column of the /proc/pid/statm file.

  PAGESIZE: obtained by running the getconf PAGESIZE command.

  MemTotal: obtained from the /proc/meminfo file.

• Collection method (Windows): Call the Windows API procGlobalMemoryStatusEx to obtain the total memory size. Call GetProcessMemoryInfo to obtain the used memory size. Use the used memory size to divide the total memory size to get the memory usage.

0–100

%

N/A

2.7.2

1 minute

specified_process_file

(Agent)Files Opened by a Process

Number of files opened by a specific process of the monitored object.

• Collection method (Linux): Run the ls -l /proc/pid/fd command to view the number of opened files.
• Collection method (Windows): Use Windows API NtQuerySystemInformation to obtain information about all opened handles in the system, check whether each handle is a file handle opened by the process, and obtain the number of files opened by the process.

  In Windows, the specified_process_file metric of some processes cannot be collected due to causes like insufficient permissions.

≥ 0

count

N/A

2.7.2

1 minute

gpu_status

(Agent) GPU Health Status

GPU health status of the VM. This metric is a composite metric.

• Possible causes: 1. The ECC exceeded the threshold. 2. The GPU memory address failed to be remapped. 3. The GPU card is in the rev ff state. 4. infoROM error. 5. There are pages to be isolated. 6. The remapped rows are incorrect. (For details, see the following detailed metrics.)
• Collection method (Linux): Call APIs from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU status.
• Collection method (Windows): Call APIs from the GPU driver library file nvml.dll to obtain the GPU status.

• 0: healthy
• 1: subhealthy
• 2: faulty

None

2.4.5

1 minute

gpu_performance_state

(Agent) Performance Status

Performance status of the GPU.

• P0-P15, P32
• P0 indicates the maximum performance status. P15 indicates the minimum performance status. P32 indicates the unknown status.
• Collection mode (Linux): Call the NvmlDeviceGetPerformanceState API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU performance level.
• Collection method (Windows): Call the NvmlDeviceGetPerformanceState API from the GPU driver library file nvml.dll to obtain the GPU performance level.

• P0-P15: P0 indicates the maximum performance status, and P15 indicates the minimum performance status.
• P32 indicates the unknown status.

None

2.4.1

1 minute

gpu_power_draw

(Agent) GPU Draw Power

Power of the GPU.

• If the power exceeds the maximum power or is an incorrect value, the GPU hardware may be faulty.
• Collection method (Linux): Call the NvmlDeviceGetPowerUsage API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU power.
• Collection method (Windows): Call the NvmlDeviceGetPowerUsage API from the GPU driver library file nvml.dll to obtain the GPU power.

≥ 0

W

2.4.5

1 minute

gpu_temperature

(Agent) GPU Temperature

Temperature of the GPU.

• If the temperature exceeds the maximum operating temperature threshold or is an incorrect value, the GPU hardware may be faulty.
• Collection method (Linux): Call the NvmlDeviceGetTemperature API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU temperature.
• Collection method (Windows): Call the NvmlDeviceGetTemperature API from the GPU driver library file nvml.dll to obtain the GPU temperature.

≥ 0

°C

2.4.5

1 minute

gpu_usage_gpu

(Agent) GPU Usage

GPU computing power usage.

• The GPU computing power usage is displayed in percentage. The value is an instantaneous value at the sampling point.
• Collection method (Linux): Call the NvmlDeviceGetUtilizationRates API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU computing power usage.
• Collection method (Windows): Call the NvmlDeviceGetUtilizationRates API from nvml.dll to obtain the GPU computing power usage.

0–100

%

2.4.1

1 minute

gpu_usage_mem

(Agent) GPU Memory Usage

GPU memory usage.

• The GPU memory usage is displayed in percentage. The value is an instantaneous value at the sampling point.
• Collection method (Linux): Call the NvmlDeviceGetUtilizationRates API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU memory usage.
• Collection method (Windows): Call the NvmlDeviceGetUtilizationRates API from nvml.dll to obtain the GPU memory usage.

0–100

%

2.4.1

1 minute

gpu_used_mem

(Agent) GPU Used Memory

Used GPU memory.

• The used GPU memory is displayed in percentage. The value is an instantaneous value at the sampling point.
• Collection method (Linux): Call the NvmlDeviceGetMemoryInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the used GPU memory.
• Collection method (Windows): Call the NvmlDeviceGetMemoryInfo API from the GPU driver library file nvml.dll to obtain the used GPU memory.

≥ 0

MB

2.4.5

1 minute

gpu_free_mem

(Agent) GPU Free Memory

Remaining GPU memory.

• The idle GPU memory data is displayed.

• Collection method (Linux): Call the NvmlDeviceGetMemoryInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the remaining GPU memory.
• Collection method (Windows): Call the NvmlDeviceGetMemoryInfo API from nvml.dll to obtain the remaining GPU memory.

≥ 0

MB

2.4.5

1 minute

gpu_usage_encoder

(Agent) Encoding Usage

GPU encoder usage.

• The GPU encoder usage is displayed in percentage. The value is an instantaneous value at the sampling point.
• Collection method (Linux): Call the NvmlDeviceGetEncoderUtilization API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU encoding capability usage.
• Collection method (Windows): Call the NvmlDeviceGetEncoderUtilization API from the GPU driver library file nvml.dll to obtain the GPU encoding capability usage.

0–100

%

2.4.5

1 minute

gpu_usage_decoder

(Agent) Decoding Usage

GPU decoder usage.

• The GPU decoder usage is displayed in percentage. The value is an instantaneous value at the sampling point.

• Collection method (Linux): Call the NvmlDeviceGetDecoderUtilization API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU decoding capability usage.
• Collection method (Windows): Call the NvmlDeviceGetDecoderUtilization API from the GPU driver library file nvml.dll to obtain the GPU decoding capability usage.

0–100

%

2.4.5

1 minute

gpu_graphics_clocks

(Agent) GPU Graphics Clocks

GPU graphics (shader) clock frequency.

• Displays the GPU clock frequencies related to graphics performance. If no graphics capability is used, you can ignore it.
• Collection method (Linux): Call the NvmlDeviceGetClockInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU graphics clock frequency.
• Collection method (Windows): Call the NvmlDeviceGetClockInfo API from the GPU driver library file nvml.dll to obtain the GPU graphics clock frequency.

≥ 0

MHz

2.4.5

1 minute

gpu_sm_clocks

(Agent) GPU SM Clocks

Streaming processor clock frequency of the GPU.

• Clock frequency for controlling the GPU memory running speed.
• Collection method (Linux): Call the NvmlDeviceGetClockInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the streaming processor clock frequency of the GPU.
• Collection method (Windows): Call the NvmlDeviceGetClockInfo API from the GPU driver library file nvml.dll to obtain the streaming processor clock frequency of the GPU.

≥ 0

MHz

2.4.5

1 minute

gpu_mem_clocks

(Agent) GPU Memory Clocks

Memory clock frequency of the GPU.

• Displays the clock frequency closely related to CUDA core computing of the GPU.
• Collection method (Linux): Call the NvmlDeviceGetClockInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the GPU memory clock frequency.
• Collection method (Windows): Call the NvmlDeviceGetClockInfo API from the GPU driver library file nvml.dll to obtain the GPU memory clock frequency.

≥ 0

MHz

2.4.5

1 minute

gpu_video_clocks

(Agent) GPU Video Clocks

Video (including codec) clock frequency of the GPU.

• Displays the codec clock frequency of the current GPU.
• Collection method (Linux): Call the NvmlDeviceGetClockInfo API from the GPU driver library file libnvidia-ml.so.1 to obtain the video clock frequency of the GPU.
• Collection method (Windows): Call the NvmlDeviceGetClockInfo API from the GPU driver library file nvml.dll to obtain the GPU video clock frequency.

≥ 0

MHz

2.4.5

1 minute

gpu_tx_throughput_pci

(Agent) GPU PCI Tx Throughput

Outbound bandwidth of the GPU.

• Displays the amount of data sent by the GPU to the host via PCIe.
• Collection method (Linux): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file libnvidia-ml.so.1 to obtain the outbound bandwidth of the GPU.
• Collection method (Windows): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file nvml.dll to obtain the outbound bandwidth of the GPU.

≥ 0

MByte/s

2.4.5

1 minute

gpu_rx_throughput_pci

(Agent) GPU PCI Rx Throughput

Inbound bandwidth of the GPU.

• Displays the amount of data sent by the host to the GPU via PCIe.
• Collection method (Linux): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file libnvidia-ml.so.1 to obtain the inbound bandwidth of the GPU.
• Collection method (Windows): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file nvml.dll to obtain the inbound bandwidth of the GPU.

≥ 0

MByte/s

2.4.5

1 minute

gpu_volatile_correctable

(Agent) Volatile Correctable ECC Errors

Number of correctable ECC errors since the GPU is reset. The value is reset to 0 each time the GPU is reset.

• Collection method (Linux): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file libnvidia-ml.so.1 to obtain the number of correctable ECC errors since the GPU is reset.
• Collection method (Windows): Call the NvmlDeviceGetPcieThroughput API from the GPU driver library file nvml.dll to obtain the number of correctable ECC errors since the GPU is reset.

≥ 0

count

2.4.5

1 minute

gpu_volatile_uncorrectable

(Agent) Volatile Uncorrectable ECC Errors

Number of uncorrectable ECC errors since the GPU is reset. The value is reset to 0 each time the GPU is reset.

• Collection method (Linux): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file libnvidia-ml.so.1 to obtain the number of uncorrectable ECC errors since the GPU is reset.
• Collection method (Windows): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file nvml.dll to obtain the number of uncorrectable ECC errors since the GPU is reset.

≥ 0

count

2.4.5

1 minute

gpu_aggregate_correctable

(Agent) Aggregate Correctable ECC Errors

Number of correctable ECC errors on the GPU.

• Collection method (Linux): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file libnvidia-ml.so.1 to obtain the number of correctable ECC errors on the GPU.
• Collection method (Windows): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file nvml.dll to obtain the number of correctable ECC errors on the GPU.

≥ 0

count

2.4.5

1 minute

gpu_aggregate_uncorrectable

(Agent) Aggregate Uncorrectable ECC Errors

Number of uncorrectable ECC errors on the GPU.

• Collection method (Linux): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file libnvidia-ml.so.1 to obtain the number of uncorrectable ECC errors on the GPU.
• Collection method (Windows): Call the NvmlDeviceGetTotalEccErrors and NvmlDeviceGetMemoryErrorCounter APIs from the GPU driver library file nvml.dll to obtain the number of uncorrectable ECC errors on the GPU.

≥ 0

count

2.4.5

1 minute

gpu_retired_page_single_bit

(Agent) Retired Page Single Bit Errors

Number of retired page single bit errors, which indicates the number of single-bit pages isolated by the GPU.

• Collection method (Linux): Call the NvmlDeviceGetRetiredPages API from the GPU driver library file libnvidia-ml.so.1 to obtain the number of single-bit pages isolated by the GPU.
• Collection method (Windows): Call the NvmlDeviceGetRetiredPages API from the GPU driver library file nvml.dll to obtain the number of single-bit pages isolated by the GPU.

≥ 0

count

2.4.5

1 minute

gpu_retired_page_double_bit

(Agent) Retired Page Double Bit Errors

Number of retired page double bit errors, which indicates the number of double-bit pages isolated by the GPU.

• Collection method (Linux): Call the NvmlDeviceGetRetiredPages API from the GPU driver library file libnvidia-ml.so.1 to obtain the number of double-bit pages isolated by the GPU.
• Collection method (Windows): Call the NvmlDeviceGetRetiredPages API from the GPU driver library file nvml.dll to obtain the number of double-bit pages isolated by the GPU.

≥ 0

count

2.4.5

1 minute

gpu_lnkcap_speed

(Agent)Max. GPU link speed

Maximum speed supported by the PCIe link of the GPU.

• Maximum data throughput capability of the GPU on the PCIe bus.
• Collection method (Linux): Use lspci -d 10de: -vv | grep -i lnkcap to query the maximum speed supported by the PCIe link of the GPU.
• Collection method (Windows): Use gwmi Win32_Bus -Filter 'DeviceID like "PCI%"').GetRelated('Win32_PnPEntity') to query the maximum speed supported by the PCIe link of the GPU.

≥ 0

GT/s

2.6.7

1 minute

gpu_lnkcap_width

(Agent)Max. GPU link width

Link width of the PCIe link.

• Maximum number of PCIe lanes supported by the GPU.
• Collection method (Linux): Use lspci -d 10de: -vv | grep -i lnksta to query the maximum speed supported by the PCIe link of the GPU.
• Collection method (Windows): Use gwmi Win32_Bus -Filter 'DeviceID like "PCI%"').GetRelated('Win32_PnPEntity') to query the maximum speed supported by the PCIe link of the GPU.

≥ 0

count

2.6.7

1 minute

gpu_lnksta_speed

(Agent)GPU link speed

PCIe connection speed of the GPU.

• Maximum PCIe link speed supported by the GPU.
• Collection method (Linux): Use lspci -d 10de: -vv | grep -i lnkcap to query the PCIe connection speed of the GPU.
• Collection method (Windows): not supported.

≥ 0

GT/s

2.6.7

1 minute

gpu_lnksta_width

(Agent)GPU link width

PCIe link width of the GPU.

• Maximum number of lanes in the PCIe link supported by the GPU.
• Collection method (Linux): Use lspci -d 10de: -vv | grep -i lnksta to query the PCIe link bandwidth of the GPU.
• Collection method (Windows): not supported.

≥ 0

count

2.6.7

1 minute

gpu_nvlink_number

(Agent)GPU NVLinks

Number of NVLink links of the GPU.

• Number of NVLink links supported by the GPU. For example, A100 supports 12 NVLink links.
• Collection method (Linux): Call the nvmlDeviceGetFieldValue API from the GPU driver library file libnvidia-ml.so.1 to obtain the number of NVLink links of the GPU.
• Collection method (Windows): not supported.

≥ 0

count

2.6.7

1 minute

gpu_nvlink_bandwidth

(Agent)Average GPU NVLink bandwidth

NVLink link width of the GPU.

• Indicates the total bandwidth for data transmission used between GPUs.
• Collection method (Linux): Call the nvmlDeviceGetFieldValue API from the GPU driver library file libnvidia-ml.so.1 to obtain the NVLink link width of the GPU.
• Collection method (Windows): not supported.

≥ 0

GB/s

2.6.7

1 minute

gpu_usage_frame_buffer_memory

(Agent) Frame Buffer Utilization

0–100

%

2.8.2

1 minute

gpu_usage_mem_rw

(Agent)GPU read/write usage

0–100

%

2.7.2

1 minute

gpu_usage_sm

(Agent) SM Utilization

0–100

%

2.8.2

1 minute