[MX] The value of 'jnxOperating1MinLoadAvg' (CPU Load Average SNMP value) is over 100

Summary:

As mentioned in KB23726 - [Junos] How to monitor Routing Engine's CPU Load Averages using SNMP, a Routing Engine's CPU load average can be monitored using SNMP in 1 minute time spans.

This article explains the value of 'jnxOperating1MinLoadAvg'. It is the CPU's load average over the last 1 minute. An example is provided, showing it as a percentage value of zero if unavailable or inapplicable. Sometimes, the value is over 100. This maybe be expected, depending on the production network environment. This value has a different meaning than the actual CPU Utilization percentage.

Symptoms:

> show system processes extensive |except 0.00%|no-more

last pid: 91449;  load averages:  0.93,  0.62,  0.45  up 482+12:22:23    17:28:46

157 processes: 7 running, 134 sleeping, 16 waiting


Mem: 552M Active, 244M Inact, 239M Wired, 1207M Cache, 69M Buf, 1261M Free

Swap: 2048M Total, 2048M Free


  PID USERNAME    THR PRI NICE   SIZE    RES STATE    TIME   WCPU COMMAND

   11 root          1 171   52     0K    12K RUN    9774.7  81.20% idle 

 2885 root          2   8  -88   122M 22232K nanslp 828.9H  6.88% chassisd

 2503 root          1  96    0   382M   328M RUN    214.6H  1.12% rpd

   20 root          1 -68 -187     0K    12K RUN    133.8H  0.93% irq10: em0 em1+++*

   14 root          1 -40 -159     0K    12K WAIT    78.5H  0.05% swi2: netisr 0


{MASTER}

> show snmp mib walk jnxOperatingEntry | match LoadAvg.9.1.0.0

jnxOperating1MinLoadAvg.9.1.0.0 = 117 

jnxOperating5MinLoadAvg.9.1.0.0 = 67

jnxOperating15MinLoadAvg.9.1.0.0 = 46


Routing Engine 0 REV 10   740-013063   9009040972        RE-S-2000

  ad0     999 MB  SILICONSYSTEMS INC 1GB C9433114328209060R07

Cause:

CPU Load Average is the same as what we see in Unix-like systems while executing utilities such as top, uptime etc. The number indicates the number of jobs in a run queue or waiting for the CPU resource, averaged out for 1 minute.

When checking this value, you also need to consider how many Cores the RE has. For example, the scenario described in this article, RE-S-2000 only has one core. So 1 minute average of 117 means the CPU was overloaded by 17%. That means 17% processes or jobs were waiting for CPU resource averaged out for the past 1 minute.

In the case of multiple core REs, 117 is still acceptable. The maximum percent capacity for a dual core processor would be 200%. Similarly, if you have a quad-core, 400% is the max value representing 100% usage.

The CPU load is different from the CPU ultilization result from the 'show system processes extensive' result.

For a better understanding on this topic, see Wikipedia's explanation on Load Computing.

Solution:

When the value of 'jnxOperating1MinLoadAvg' is more than 100, it does not mean the current CPU ultilization is over 100%. This is expected depending on the production environment.

To check the CPU utilization as a percentage, use 'jnxOperatingCPU' :

> show snmp mib walk jnxOperatingCPU |match 9.1.0.0

jnxOperatingCPU.9.1.0.0 = 3

Knowledge Base