Zabbix - Oracle performance
Hi,
In a last month Zabbix write cache % free have droped from ~98-99% to 70%. When server is restarted is cleaned and it slowly fills cache again (~1h)...
As far I know it should be performance DB related issue (In our case - Oracle) - or I'm wrong?
We found some I/O problems on our dedicated DB server and moved DB to more faster disk array. Now when problems with I/O is solved there is still 70% free cache. No changes... Disk controller works with about ~10% workload.
If we look at queue there is no items waiting more than 10 sec. So it seems to be fine.
Still seems there is some Zabbix application related issues which our Oracle DBA pointed out - "SQL*Net message from client" has a long waiting time. Which means that Zabbix application can't send data?!
Quote from oracle forum:
To solve this problem I tried to increase amount of DB syncers - at the beginning it was 15, now it's changed to 64. Chaning this parametr decreased "SQL*Net message from client" 3x times from 150 000 000 to 46 000 000, but still it did no affect write cache.
Zabbix status:
Number of hosts: 483
Number of items: 23673
Number of triggers: 2930
Required server performance, new values per second: 178.02
Zabbix server:
Centos 5.5, 8GB RAM, RAID 10, 1Gbps network
DB (Oracle 10):
AIX 6.1, 16GB RAM, RAID 10 (SAN disk array), 1Gbps network
So my question is - is write cache should be at level of 70% free, isn't that indicating problems in performance? If it's indication of performance problems what should I change to solve them?
In a last month Zabbix write cache % free have droped from ~98-99% to 70%. When server is restarted is cleaned and it slowly fills cache again (~1h)...
As far I know it should be performance DB related issue (In our case - Oracle) - or I'm wrong?
We found some I/O problems on our dedicated DB server and moved DB to more faster disk array. Now when problems with I/O is solved there is still 70% free cache. No changes... Disk controller works with about ~10% workload.
If we look at queue there is no items waiting more than 10 sec. So it seems to be fine.
Still seems there is some Zabbix application related issues which our Oracle DBA pointed out - "SQL*Net message from client" has a long waiting time. Which means that Zabbix application can't send data?!
Quote from oracle forum:
sqlnet message from client is typically not an indication of the database having a problem, but rather the client application not asking us to do anything!
To solve this problem I tried to increase amount of DB syncers - at the beginning it was 15, now it's changed to 64. Chaning this parametr decreased "SQL*Net message from client" 3x times from 150 000 000 to 46 000 000, but still it did no affect write cache.
Zabbix status:
Number of hosts: 483
Number of items: 23673
Number of triggers: 2930
Required server performance, new values per second: 178.02
Zabbix server:
Centos 5.5, 8GB RAM, RAID 10, 1Gbps network
DB (Oracle 10):
AIX 6.1, 16GB RAM, RAID 10 (SAN disk array), 1Gbps network
So my question is - is write cache should be at level of 70% free, isn't that indicating problems in performance? If it's indication of performance problems what should I change to solve them?
Attached Files
Comment