Understanding default SNMP utilization calculations

Metric Name: kentik.snmp.CPU

CPU is generally returned in a direct OID that provides a integer or float value representing percentage utilization. In rare cases, there is only a result for CPU idle, which is translated to CPU using this formula:

CPU = 100 - CPU Idle

Metric Name: kentik.snmp.MemoryUtilization

Unlike CPU, memory utilization is rarely presented as a direct OID value. To calculate the percent utilization, ktranslate uses this logic:

If Memory Used and Memory Free are available:
  Memory Utilization = ( Memory Used / (Memory Free + Memory Used) ) * 100

If Memory Total and Memory Free are available:
  Memory Utilization = ( (Memory Total - Memory Free) / Memory Total ) * 100

If Memory Total and Memory Used are available:
  Memory Utilization = ( Memory Used / Memory Total ) * 100

If Memory Total, Memory Buffer, and Memory Cache are available:
  Memory Utilization = ( ( Memory Total - (Memory Buffer + Memory Cache ) ) / Memory Total ) * 100

Metric Name: kentik.snmp.IfInUtilization | kentik.snmp.IfOutUtilization

Interface utilization follows the industry standard approach of calculating the delta in bits and dividing by the product of the interface's configured speed and the time delta since the last collection was made.

For example, assuming 1 is the previous data point and 2 is the most recent:

( ( ifHCInOctets_2 - ifHCInOctets_1 ) * 8 * 100 ) / ( (sysUptime_2 - sysUptime_1) * ifSpeed )

Translated calculation

For the numerator:

• Take the latest count of octets and subtract the value of the previous sample to get a delta.

• Multiply the result by 8 to convert octets to bits.

• Multiply the bits by 100 to set up the percentage calculation.

  For the denominator:

• Take the latest value of sysUptime and subtract the value of the previous sample to get a delta that shows the polling interval.

• Multiply the interval by the configured ifSpeed (which is reported in bits).

  Finally:

• Divide the numerator by the denominator to find the utilization percent.

  Source data

  KTranslate, uses the value of either ifHCInOctets (receive) or ifHCOutOctets (transmit); replacing ifSpeed in the denominator with the value of ifHighSpeed as needed:

  Dica

  A common reason for seeing inaccurate interface utilization percentages is the configured interface speed on the device doesn't reflect the real interface speed. For instance, a 1GB MPLS circuit on a 10GB interface would show percentages at only 10% of the real utilization. To resolve this, consult your vendor's documentation on setting the interface bandwidth.

Metric Name: kentik.snmp.ifHCInOctets | kentik.snmp.ifHCOutOctets

Interface throughput uses the value of either ifHCInOctets (receive) or ifHCOutOctets (transmit).

It's important to note exactly what you are looking for when querying for throughput.

• 1 octet = 1 byte

• 1 octet = 8 bits

  The raw measurement in SNMP is counting octets, which in this context would be synonymous with bytes. You can easily calculate bytes directly through the reported metric without any conversion needed.

  If you are looking for bits per second (bps), you need to both multiply the octets and convert the final number into a 'per second' rate.

  Example

  FROM Metric SELECT
  max(kentik.snmp.ifHCInOctets) AS 'total_received_bytes',
  max(kentik.snmp.ifHCOutOctets) AS 'total_sent_bytes',
  rate(average(kentik.snmp.ifHCInOctets)*8, 1 SECOND) AS 'average_received_bps',
  rate(average(kentik.snmp.ifHCOutOctets)*8, 1 SECOND) AS 'average_sent_bps'
  FACET entity.name, if_interface_name
  SINCE 1 WEEK AGO
  TIMESERIES

  Copiar

Metric Name: kentik.snmp.ifInErrorPercent | kentik.snmp.ifOutErrorPercent

Interface error percentage uses the value of either ifInErrors (receive) or ifOutErrors (transmit), divided by either ifHCInUcastPkts (receive) or ifHCOutUcastPkts (transmit). In ktranslate, the formula looks like this:

( ifInErrors / ifHCInUcastPkts ) * 100
or
( ifOutErrors / ifHCOutUcastPkts ) * 100

Metric Name: Various

Other SNMP metrics are converted based on the existence of the enum and conversion functions in their respective SNMP profile.