We have a cluster with physical machines HW

Each machine contains 32 CPU

# lscpu
Architecture:          x86_64
CPU op-mode(s):        32-bit, 64-bit
Byte Order:            Little Endian
CPU(s):                32
On-line CPU(s) list:   0-31
Thread(s) per core:    2
Core(s) per socket:    8

Now we want to send an alarm each time the load average from uptime to 5 min from uptime reaches the threshold

It is my understanding that we need to worry when the load average (at 5 min from uptime) value is more than 32 ( number of CPUs )

Am I correct ?


The load average does not primarily show the CPU load, it is an approximation on the length of the work item queue. In real life this is often influenced much more by the CPU cycles wasted in I/O waits, then in CPU cycles actually used (ofcourse this is workload dependent).

Take a look a the top tool to determine a first estimate on how your CPUs spend their time. As a rule of thumb (and only as such) consider

  • High User percentages are healthy - your machine is in good use
  • High Sys percentages might point to a not very efficient application (or VM guests running a not optimized OS/driver combination)
  • High Wait times mean your I/O (most likely storage) can't keep up. This can again point to the need for optimization inside your application or just be asking too much from your storage hardware (Top places to look: Are rotating disks used for latency-sensitive workloads? Are you using RAID5 for small writes?)
  • we have hadoop cluster ( HDFS nodes ) , and I see that THP is enables when hadoop recommended to disabled , do you think this is the problem ? – King David Mar 19 at 23:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.