Let me illustrate:
informix@freebnort:/home/informix
pts/4:ksh$ sar 5 12
Linux 2.6.16.42-0.10-smp (freebnort)     07/08/08
22:25:12          CPU    %user    %nice    %sys    %iowait     %idle
22:25:17          all    10.66     0.00      12.52       0.00     76.83
22:25:22          all 9.13     0.00      12.30       0.00     78.57
22:25:27          all    12.04     0.00      12.76       0.00     75.20
22:25:32          all    11.46     0.00      12.69       0.00     75.85
22:25:37          all    10.14     0.00      12.69       0.00     77.17
22:25:42          all    12.81     0.00      13.17       0.00     74.02
22:25:47          all    12.15     0.00      12.71       0.00     75.14
22:25:52          all    13.76     0.00      13.10       0.00     73.14
22:25:57          all    21.79     0.00      12.89       0.00     65.32
22:26:02          all    14.84     0.00      12.64       0.00     72.53
22:26:07          all    12.11     0.00      12.47       0.00     75.42
22:26:12          all    13.43     0.00      13.18       0.00     73.38
Average:          all    12.86     0.00      12.76       0.00     74.38
informix@freebnort:/home/informix
pts/4:ksh$
(Pardon the formatting, blogger is doing my head in on this.)
As you can see, the %sys is the same order of magnitude as the %user. Given that this server is only running IDS, something is wrong. The %user should be at least an order of magnitude greater than than %sys. In general, if a database server is only a database server, then if %user is 10%, then %sys should be about 1%. If the server has a lot of connections, this might get a bit higher, but %sys shouldn't generally be more than about 3% if %user is about 10%.
Things to watch out for are:
- Swapping, especially if %sys is significantly greater than %user and the server doesn't seem to be doing anything.
- Very high rates of network activity, such as lots of connect / disconnect activity; or poorly configured connectivity, leading to excessive transmission of small packets, rather than fewer transmissions of larger packets.
Obviously, there could be many other causes, but these are the most common.
 
 
1 comment:
Thanks for the tip!
Post a Comment