System Administration
cancel
Showing results for 
Search instead for 
Did you mean: 

SO hp-ux 11.11 disks with high %busy times in glance

cartman_2
Frequent Advisor

SO hp-ux 11.11 disks with high %busy times in glance

Hi,

We have a Server with the SO disks, internal scsi disks in a rp8420, that spend lots of time at 100% busy, the disk queue gets into the 20's

we are getting this data through glance.

A couple of questions:

I have other servers, that don't have that much io on the SO disks, how can I track/look for the processes causing this IO?

Can this IO queueing on the SO disks have a negative performance impact on the oracle databases running on the server, or are they completely independent?
8 REPLIES
Elmar P. Kolkman
Honored Contributor

Re: SO hp-ux 11.11 disks with high %busy times in glance

A good indication can be the top I/O process when selecting the filesystems-list ('i' in glance).
Every problem has at least one solution. Only some solutions are harder to find.
cartman_2
Frequent Advisor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Yes, the problem is this server has loads of IO from several databases, and I use xglance and sort processes by IO rate, but I can't manage to associate processos with which disks/fs they are writing to, to the SO disks ti should some kind of root processes but I can't see any process from root with high IO rates
muruganantham raju
Valued Contributor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Hi,
You can also use the following glance adviser file to find out the process that perform heavy IO.

#cat sample_adviser
print "GBL_STATTIME ", GBL_STATTIME
proc_io_rate=0
proc_phys_read=0
proc_phys_write=0
proc loop
{
print "PROC ID", PROC_PROC_ID, " PROC_DISK_PHYS_IO_RATE ", PROC_DISK_PHYS_IO_RATE, " PROC_DISK_PHYS_READ ", PROC_DISK_PHYS_READ, " PROC_DISK_PHYS_WRITE ", PROC_DISK_PHYS_WRITE
if ( PROC_DISK_PHYS_IO_RATE > proc_io_rate ) then
{
proc_io_rate=PROC_DISK_PHYS_IO_RATE
proc_id1=PROC_PROC_ID
}
if ( PROC_DISK_PHYS_READ > proc_phys_read ) then
{
proc_phys_read=PROC_DISK_PHYS_READ
proc_id2=PROC_PROC_ID
}
if ( PROC_DISK_PHYS_WRITE > proc_phys_write ) then
{
proc_phys_write=PROC_DISK_PHYS_WRITE
proc_id3=PROC_PROC_ID
}
}
print proc_id1, " has performed highest IO RATE ", proc_io_rate
print proc_id2, " has performed highest Physc Read ", proc_phys_read
print proc_id3, " has performed highest physc Write ", proc_phys_write
print "---------------------------------------"

#/opt/perf/bin/glance -aos ./sample_adviser

Can this IO queueing on the SO disks have a negative performance impact on the oracle databases running on the server, or are they completely independent?
Yes, it will impact the performace. If the Oracle DB processes are waiting in IO queue, it will definitely affect the response time.
HTH
Muru
chris huys_4
Honored Contributor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Hi,

Where does "SO" stands for in "SO disks" ?

Greetz,
Chris
cartman_2
Frequent Advisor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Muru, thnx for the glance script, but as I said I can't find any root process doing loads of IO.

Cris, I don't understand your question.. ?
Ganesh Balraman
Regular Advisor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Hi !

Performance should be monitored in below procedure.

1.Check if both vhand and swapper are active using glance

2.use "w" option in glance and check swap usage

3.use swapinfo -mta to get more details on swap

4.use sar -b and check read cahche and write cache ratio

5.use sar -v to check inode usage

6.use sar -u to find cpu load average

7.use glance global report anc check cpu load average i.e should not be more than 3.



After all this checking comes to disk

1.sar -d "find out which disk is heavily loaded

2.use "d " in glance also to get more details.Cjeck Q length"not more than" 3 and UTIl values"NOt more than 50"

3.Find out whcih LV does disk belongs to.

MOnitor is is just spike or get prolonged.


Dennis Handly
Acclaimed Contributor

Re: SO hp-ux 11.11 disks with high %busy times in glance

>I don't understand your question?

What does the acronym SO mean?
>>What does "SO" stand for in "SO disks"?
cartman_2
Frequent Advisor

Re: SO hp-ux 11.11 disks with high %busy times in glance

Yes, As I said, the %busy of the OS disks(vg00) is every now and then at 100%, and the disk queue goes up to 20/25, I have oracles binaries in a lvol from vg00, the rest of the database(datafiles,archives,redo,etc) are all in san disks that have no io problem.

My question is:

¿Can this io botleneck in the OS(vg00) disks(mirror) have a negative performance impact on the oracle database?

Thnx