For the last month I have been working on ASM migration project for one of our client in United States.
lient wanted to move couple of environments (including Production, QA, and Dev/Patch) to ASM, critical production environment running EBusiness Suit 188.8.131.52 on raw devices utilizing 5.1 TB storage, other instances are utilizing cooked file systems.
I really enjoy working with ASM, the day it was released I got passionate about this feature and have been playing in my lab. My manager considered my interest with ASM and assigned me to this project (thanks to him). I am very much excited for this project, let’s see how this goes.
This project includes OATM (Oracle Application Tablespace Model) conversion, re-write solution for Cluster Failover (HP Service Guard is configured for Cluster Failover), DR is implemented using EMC Mirror View, BCV & backup methodology.
I would like share my experience after first migration to ASM. This migration was done using fndtsmig.pl (Oracle given utility to migrate segments). This utility migrate (move/rebuild) segments based on pre-defined rules to choose appropriate tablespaces.
We just finished with our first Iteration, I was more interested to see equal distribution of I/O across the Luns. I pulled below report from 10g grid control for %usage of one diskgroup.
It shows well balanced equal usage distribution (it’s awesome, isn’t it),
let’s have a look for IO usage
As soon as I saw this page, I noticed 3 things.
1- IO distributions are quit close/equal for all the Luns of DGOAPSFC. Other columns Total Read Calls & Total write Calls are almost equal for all the Luns. That is why I like ASM for this capability.
2- BUT this is not the case with DGOLTPFC diskgroup , this is really strange that 4 disks out of 10 Luns got 40% more high I/O.
3- I noticed there is huge difference in IO response time for 2 disks in DGOLTPFC
Point no. 1 is quite obvious and expected, for second point I tried to find the cause but still this is an open issue for me. I am working on this and would post more if I could get something.
For 3rd point, I immediately made call to my storage administrator and asked for the same if something wired at storage end for specifically for 2 disks as there is huge difference in IO response time. Other disks avg. response time is near 5.5 ms where as LUN 43 and LUN 46 it’s more 9 ms.
Storage guy asked me to give some time, after some time he told me that because of space constraint he could not give me all FC disks, those 2 disks are SATA. J
For final implementation we would be provided FC and SSD disks only so we started using descriptive naming convention.
Again, I am thrilled to work in this project and would share my experience.