onstat -P

[jklerp]

Hi,

We've got some performance problems with a baan/informix installation.
After informix restart, the application is quite fast for few days.
But after some days buffer wait (and io wait at os-level) starts to increase.


I found a health_check script (oninit.com).
It gave me warning:
Buffers allocated to index data is dangerously high [97.68%].

The script took that from onstat -P :
-------

# onstat -
Informix Dynamic Server Version 7.31.FC6 -- On-Line -- Up 12 days 03:26:28 -- 4641784 Kbytes
#onstat -P | tail -4
Data 1.69
Btree 97.68
Other 0.63

---------------

Is this bad?
Sounds quite odd to me that indexes are using all the buffers.







/Vesa

 

[cdlvj]

Is the app doing a lot of inserts?

Many index's on a table?

 

[jklerp]

The answer is yes for both questions.

There are lot of inserts and there are also quite a lot indexes!

 

[cdlvj]

Are these batch?

There is a FillFactor parm which could be adjusted to give you some help. Default I think is 90% meaning when you get a lot of inserts, this becomes full and has to start its overflow logic involving more overhead.

Look at how often commits are done.

What about the possibility of re-org of that table during off peak hours, an Unload and Reload.