1747971 Members
3551 Online
108756 Solutions
New Discussion

Re: Bottlenecks galore

 
Dave Chamberlin_1
Occasional Contributor

Bottlenecks galore

I am the new unix admin here and have an hp9000/k460
server w/4 CPUs. I have been monitoring the system
for a couple weeks now using glanceplus and top. These
have indicated many bottleneck problems and I would
like to know if someone can comment on these.

1 - CPU bottlenecks. The system only has one user,
namely ORACLE. There are commonly 100-200 Oracle
processes running and there are periods for an hour or more where the
total CPU utilization is pegged around 100%. Most of the day it is over 80%. Is
this telling me that our system is underpowered?

2 - There are frequent disk bottlenecks indicated. There are 8 disks with
mirrors (at least there are 16 different hardware addresses and only 8 logical
volumes). I have noticed that the volumes and their mirrors are on the same
controllers. I have never used a system with mirroring - is this normal? I
thought one would want them on separate controllers to break the load up. In
the system, one controller seems to be used for 8 devices, two which are having
the heaviest loads. Can I improve disk performance by shuffling things around?

3 - There have been frequent memory bottlenecks. I am not sure what the best
metrics to monitor are. At the moment I am monitoring pageouts, page faults and
page requests. Are there better indicators? Should I be more interested in
rates or actual numbers?

Any comments would be appreciated.

Thanks

 

 

p.S. This thread has been moevd from HP 9000 to HP-UX > sysadmin. - hp forum moderator

1 REPLY 1
Randy Mather
Frequent Advisor

Re: Bottlenecks galore

1 - CPU bottlenecks. The system only has one user,
namely ORACLE. There are commonly 100-200 Oracle
processes running and there are periods for an hour or more where the total CPU
utilization is pegged around 100%. Most of the day it is over 80%. Is this
telling me that our system is underpowered?
(response)
Oracle is a CPU hog. What is the run queue(load average). What application are
you running ? You are probably pegged on CPU's.

2 - There are frequent disk bottlenecks indicated. There are 8 disks with
mirrors (at least there are 16 different hardware addresses and only 8 logical
volumes). I have noticed that the volumes and their mirrors are on the same
controllers. I have never used a system with mirroring - is this normal? I
thought one would want them on separate controllers to break the load up. In
the system, one controller seems to be used for 8 devices, two which are having
the heaviest loads. Can I improve disk performance by shuffling things around?
(response)
Spread out the disks on other controllers. Put the Oracle index on a seperate
disk and controller if possible. Which disk array(s) are being used ?

3 - There have been frequent memory bottlenecks. I am not sure what the best
metrics to monitor are. At the moment I am monitoring pageouts, page faults and
page requests. Are there better indicators? Should I be more interested in
rates or actual numbers?
(response)
Look at how much swap is configured. How much memory each process is using (an
average). Check the size of the buffer cache. (look at kernel parms dbc_min_pct
and dbc_max_pct) set the to a constant 7-10% of physical memory.

Performance problems are not easily fixed overnight, you need to factor in
other variables.

You are welcome to email if you wish.

randy_mather@ham.honda.com



--------------------------------------------------------------------------------
Thread:
Bottlenecks galore Dave Chamberlin 04/27/2000 08:21:01 AM *you are here*


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

Click here to share this discussion thread with a friend....