Ad Widget

Collapse

Zabbix Performance Tuning

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • jdzbx
    Junior Member
    • Dec 2015
    • 4

    #1

    Zabbix Performance Tuning

    Running Zabbix 3.0 processing 186 values per second. CentOS 7 VM with 2 vCPU and 8GB RAM and MariaDB for storage.

    I'm seeing a high amount of disk I/O from this vm and trying to determine if there is any performance tuning I can do to help reduce it. Obviously the I/O is from MariaDB.

    I'm seeing about 1400 writes/sec and around 40-50MB/s of disk throughput. Does this seem normal for the amount of values per second that zabbix is processing?

    Here's a screenshot of the disk I/O from the zabbix vm: http://i.imgur.com/MWv3G1n.png
  • LenR
    Senior Member
    • Sep 2009
    • 1005

    #2
    You might run one of the mysql tuning scripts, like http://www.askapache.com/mysql/perfo...ing-mysql.html

    Comment

    • Stefan_D
      Junior Member
      • Sep 2015
      • 18

      #3
      How did you set up your Database File System?
      Does the DB have it's own FS?
      If yes, did you mount it with "noatime" option?

      --
      SD

      Comment

      • Colttt
        Senior Member
        Zabbix Certified Specialist
        • Mar 2009
        • 878

        #4
        also a good solution is to set the Linux-IO-scheduler to nnop

        http://www.cyberciti.biz/faq/linux-c...-for-harddisk/
        Debian-User

        Sorry for my bad english

        Comment

        • jdzbx
          Junior Member
          • Dec 2015
          • 4

          #5
          Thanks for the feedback. I had added a number of network devices, each of which had all of their interfaces enumerated. There were approximately 3500 items per device. I added a filter to remove all of the unnecessary interfaces from the LLD.

          After doing so, housekeeping was unable to run successfully. The number of items in the housekeeping table would never go down and the housekeeper busy % was 100. I changed the number of housekeeping items from the default 5,000 to 100 and it was able to complete an interval. After about 8-10 hours it had removed all of the old data from the removed devices and disk I/O went down to an average of around 20-30 iops.

          Comment

          Working...