10.20 to 11.0

[ben_43]

Team:

We are upgrading the T500 servers running oracle 8.05 from 10.20 to 11.0. Please let me know what the known issues are with this kind of upgrade. Also this is my first upgrade. Please help.

Thanks
Ben J

[Patrick Wallek]

There have been LOTS and LOTS of discussions on this topic in these forums. Your best bet is to click on 'search' on the upper left and put in "10.20 11 upgrade" as your keywords. Be sure to select "ITRC Forums" to search in. You will get LOTS of results.

I just tried this and the search did work and I got lots of results.

Here are 3 of the links I found on this topic:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x33733fd7f72ad611abd50090277a778c,00.html

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xa6ecf715edc6d5118ff10090279cd0f9,00.html

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xb704a24d9abcd4118fef0090279cd0f9,00.html

[Martin Johnson]

I would recommend a cold install of 11.0 rather than an upgrade. I upgraded 4 systems from 10.20 to 11.0. None of them were successful the first time around. When they were successful, they took 24 to 36 hours to complete.

My system disk became unbottbale during the upgrade due to the fact that installing Mirror-UX on v11.0 reinstalls the LVM software without warning. The new LVM software is *NOT* patched. All LVM patches need to be re-installed prior to doing any LVM activity or the system disk becomes unbootable.

Rather than go through that pain again for my other systems. I ignited one of my v11.0 boxes and cloned rest of 10.20 to 11.0. Much faster (less than 4 hours).

HTH
Marty

[Dave Chamberlin]

Cold install is recommended. I have done cold installs for 4 systems. None had any install related problems. My installs took about 8 hours to get things how I wanted. The 11 software uses more disk space - take the opportunity to proactively optimize your var,usr, partitions etc. I would suggest a full backup before you start. Make an extra copy of config files (like etc/hosts, etc) on another host for reference or easy access. Make sure you know where your license codes are for things like GlancePlus, Mirror-ux etc before you start. Make sure you read the release notes etc. There are some disk firmware issues etc that might be an issue for you.

[S.K. Chan]

I do not forsee much problem from the OS side and you've already been given the links to help you decice cold-os-install vs upgrade-os. No matter what my preference has always been doing cold-install. The challenge you're going to face is getting Oracle to perform well on the new OS (11.0) compare to 10.20. I guess that is the issue you wanted the answer for ? I'm not an Oracle person but I would watch out for 2 things ..
- Patches (especially those required by Oracle)
- The recommended kernel parameters setting for Oracle on 11.0.
You may search the forum to get as much info for these 2 items or post a separate thread to get the answers.

[ben_43]

Thanks. The oracle point was of prime importance. There is no question of cold-install as the management have decided on the upgrade. Given this case what would be the considerations and also the oracle Kernel parameters and Patches.

Thanks Again,
Ben

[Chris Wilshaw]

My company are currently going through the migration process too (fortunately for us, we've been able to go for the cold install).

A couple of issues that we have come up against

1) mib2agt process can consume large amounts of memory (PHSS_21046 fixes this).

2) screen performance can be seen to be slower on 11.* than 10.20 - adding the -TCP_DELAY option to the telnetd daemon in /etc/inetd.conf can resolve this (alternatively you can use the -s and -z parameters to set your own buffer and buffer timeout values).

Once you've completed an upgrade, make sure that your company has got testers available from various business areas - this takes the pressure off you to do all the testing. Make sure they test every possible aspect of the system (especially if you have any extra bits of hardware - scanners for example that connect to it).

[ben_43]

Thanks Chris:

Can you explain your second point a little more about all those options please.

Thanks
Ben.

[Chris Wilshaw · Accepted solution]

Certainly.

With the move from 10.20 to 11.*, new functionality has been added to the telnetd process, which can have an impact on the performance of any connections using it (PC's, dumb terminals etc).

the -TCP_DELAY option allows small data packets to be combined, prior to transmission (which can in some instances lead to performance enhancements).

the -s and -z options are alternatives to -TCP_DELAY, which allow you to set your own values for the packet size (in bytes) to transfer and the time span (in CPU clock ticks) before the data is transmitted.

Neither of the -s or -z options have a default value, and it appears that it is a case of trial and error to find the optimal level for your particular application. In our case, we have -s set to 400, and -z to 5

The options (whichever is best suited) need to be added to the end of the line

telnet stream tcp nowait root /usr/lbin/telnetd telnetd

in /etc/inetd.conf, and then the inetd process itself restarted by running

inetd -c

Obviously, any performance differences are also subject to the speed of your network connections.