Performance Issue in XP12000 with rp8420

[Arun Jain]

Dear All

I have one rp8420 box containing 2 npars. One is running Database (Oracle) and second is running Application.
System Configurations of each server :
OP SYSTEM: HP-UX B.11.23 64 bit (both)
NO CPUs: 10 (both)
CPU SPEED: 1100 MHZ
Main Memory : 40 GB and 48 GB (one each)

XP12000 Storage :
Database server - 120 assigned LUNs + 2 local HDD.
Each LUN contains 19.6 GB and RAID 1
OS Level : Striped with 8 HDDs
Each Logical Volume contains combination of 8 LUNs and size is approx 153 GB each.

Application server - 2 local HDD

Issue : Server Performance Issue
Average System Load is reaching maximum upto 5.00
Disk I/O wait is high.
Kernel Parameter : scsi_max_qdepth value is 8 (default)
scsi_maxphys : 1048576

I am attaching some logs for the reference in a file with the thread.

Please suggest :

Thanks & Regards :
Arun Jain

[Duncan Edmonstone]

Arun,

The avg service time for the XP LUNs seems a little high - are these LUNs in any sort of Continuous Access relationship (ie replicated to another XP)

HTH

Duncan

[Arun Jain]

Duncan,

Yes, your guess is right. These LUNs are continuously replicating on a DR site 24x7 on a XP12000 Box.

Arun

[Steven E. Protter]

Shalom Arun,

You might want to stop the continuous replication and do it once a day during a time when system use is lower.

Load factor is not really meaningful here, I/O wait is.

Performance measurement tools:

http://www.hpux.ws/?p=6

SEP

[Duncan Edmonstone]

Arun,

Presumably you are replicating synchronously?

If that's the case, then chances are you can't do anything about IO service times as each write IO has to be written into both XPs cache before the write IO can be acknowledged. You could consider moving to asynchronous or journaled replication, but there's probably be a cost.

If you're already operating asynchronous or journalled replication, there are other things that can be done to the XP to speed things up (more MPs, more cache, more journal disks etc.), but again there'll probably be a cost.

What's the distance between the 2 XPs? What's the bandwidth/latency of the link bewtween the XPs? How was the link and the XPs sized in the first place?

HTH

Duncan

[sreekanthtm]

This is a complex scenario. Better to log a call in HP, and get assistance from there.
Performance issue can be due to,
1. Fabric problem.
2. No: HBA's used to connect to XP.
3. Need to verify switch side configuration.
4. Server performance ( processor, RAM, n/w parameters)
5. XP configuration (No: of MP's used to present the LUN, Is it properly configured for load balancing? Is the same MP is sharing any other host? etc..)
6.History of the server, XP and the fabric. ( Is there any recent change happened? If yes needs to analyze closely.)
7. What abt the load balancing from Server side? ( Are you using secure path? If yes, what is the mode?)

So, better to get assistance from HP.


Best wishes
Sreekanth

[Suraj K Sankari]

Hi,

Above all what is your syslog.log saying? Did you check your syslog.log any hardware related or ibolt error is there check your syslog.log.


Suraj