Upgrade from 8.10 to 9.0

Sanjiv Garg

Active Member
Hello All,

Our company is planning to upgrade JDE 8.10 to 9.0. Following is the detail:

Current Configuration:

JDE Release: 8.10/8.96.3.1

E1 Servers: 3x PA-RISC HP-UX 11.11. First server is PD Batch, second is PD App and third one is DV/PY App+Batch

Database Servers: 2x PA-RISC HP-UX 11.11. One for PD database also holding System Path Code and second for PY/DV database.

Total Users: 600.

Concurrent Users: 150.

Our plan is to upgrade JDE to 9.0/8.98x. Our current version of UNIX (11v1) is not supported with 8.98. So we have to perform OS upgrade before JDE/Tool upgrade. Our lease to HP PA-RISC is also expiring in 2010. So it makes sense to upgrade the hardware along with JDE upgrade. PA-RISC is end of life product and we have no option other than to move different platform.

First choice is to move to HP Itanium platform. Database migration to Itanium from PA-RISC looks straight forward. But our current release of JDE (8.10) is not supported on Itanium. So not sure what would be the upgrade strategy…

Second thought is to move existing E1 servers to Windows environment and leave database running on UNIX. I believe database migration from UNIX to Windows require lot more planning and resources as compared to moving just E1 servers. In this plan, we think to add Windows E1 server in current release and later perform the JDE upgrade. So, again not sure how this approach would be…

Also is there any issues running E1 on Windows server for that many users? I mean performance issues? Since v9.0 is thin client release, so a lot more processing would be done by E1 servers.

I appreciate to get your feed back and also help to plan out this upgrade.

Thanks,

Sanjiv
 
Many customers are migrating away from the PA-RISC solution because of the end-of-life issues in 2010. Some customers are migrating to Intel Linux, some are migrating to Solaris or AIX and others are migrating to HP Itanium. Whichever solution you want to migrate to, the steps are the same.

First of all, you should absolutely try to migrate your platform prior to the upgrade. Its possible to do both in the upgrade process - but theres more risk because there are two items being changed instead of one.

First of all, decide on the platform you want to go to. Then migrate everything to that platform. If you want to move to Windows - then you can migrate to windows with Oracle on Windows, and there isn't much that has to change at all. Only if you were going to perform a database migration to, say, SQL Server, would there be the need to plan for migration tasks.

However, changing platforms as far as OneWorld is concerned is slam dunk. As long as you have the CD's or tools release/ platform pack available on the exact version you're running (8.96) for the platform you're going to - then Oneworld is easy to migrate - in fact, you'd probably just ADD the servers into your existing environment, and make changes to OCM mappings etc to achieve. Very little downtime if any.

Outside of OneWorld is a completely different story though. If you have 3rd party applications coded specifically to the HP=UX platform, then those will be a challenge to migrate. You need to inventory those applications carefully and test their counterparts prior to your migration.

Lots and lots of testing is paramount - but in your situation, you need to eliminate the PA-RISC HP/UX as much as possible and migrate to a platform that is better supported for an ongoing basis.

As for running numbers of users on Windows - I have customers with 600+ concurrent users on windows platform with NO issues - and thats with SQL Server. There really is absolutely no issues with running any size numbers of users on any of the platforms supported under E1 anymore. SQL Server is just as scalable as, say, Oracle 8i or 9i was - and is certainly as scalable as iSeries. I'd say that if you were planning on implementing more than 1,000 concurrent users, then you might need some specialist architecture to ensure support of large numbers of transactions - but at the levels you're running, its a slam dunk.

Hope that helps.
 
Sanjiv,

5 Years ago we switched from HP PA-RISC to a 32bit Windows platform for both JDE and Oracle DB. The platform switch itself was a relatively minor part of our upgrade from XE to 8.9. Today our Oracle DB is on Windoze Server 2008 - 64bit. FYI we have ~ 150 concurrent users.

First - carefully evaluate whether you want to go with the Itanium chipset - its really a niche product and you very well could find yourself in the same situation you're in now with the PA-RISC. If you want to stay with UNIX take a good look at IBM's offerings.

Second - regarding database migration. Oracle DB is very easy to migrate across platforms from an application perspective - its really almost invisible to the applications whether you move an app from UNIX to Windows or vice versa. In my opinion moving the database to a different platform is easier than moving the JDE servers. Your main concern will be the change in backup software to use. With JDE support for 64bit Windoze the windows alternative is very viable now for sites with large #'s of users. Despite all that Oracle DB DOES run very well on UNIX - you can't go wrong with that decision.

Third - regarding # of users on a windows server. The nice thing about this architecture is that you can easily add very cheap wintel boxes if you underestimated your hardware requirements.

Fourth - where you really need to spend your hardware bucks is not on the servers but on your secondary storage. A really good SAN with a RAID 1+0 / 10 configuration will do far more for overall performance than a high powered server using a basic RAID-5 setup.

There's my opinion for what its worth.
 
[ QUOTE ]
Fourth - where you really need to spend your hardware bucks is not on the servers but on your secondary storage. A really good SAN with a RAID 1+0 / 10 configuration will do far more for overall performance than a high powered server using a basic RAID-5 setup.



[/ QUOTE ]

I totally agree with Larry's last point. We've gone through three generations of SANs for our JDE servers. I have seen my package builds build time decrease from six hours to around 45 minutes while we have stayed on the XE release. Fast servers help, but a fast SAN is a huge help in increasing performance.

- Gregg
 
Thanks a lot Jon!

This information is very handy to draw our project plans.

Based on your feed back, we may go with one of the following options:

1. Migrate all E1 servers including Oracle database to HP UX Itanium platform.

2. Migrate all E1 servers to Windows Intel platform and migrate Oracle database over to HP UX Itanium platform.

If you can answer couple of more questions Jon, that would be muchly appreciated:

1. If we choose to migrate to Windows Intel platform for E1 servers (excluding database server), how much consultant work is required given that base infrastructre is in place?

2. We will go through vendor sizing exercise, but wondering how many Windows E1 servers do you think we need to handle 150+ concurrent keeping in mind that an upgrade to v9.0 will follow platform migration which is 100% thin client solution ?

3. At present we use HP Service Guard for failover for E1 servers, migration to Windows platform require re-visiting failover solution. We may need some loadbalancing solution as well as JDE next release is 100% thin client. What would be the best solution for Failover/Loadbalance.

Thanks again for your input.

Sanjiv
 
Hi Sanjiv

[ QUOTE ]

1. Migrate all E1 servers including Oracle database to HP UX Itanium platform.

[/ QUOTE ]

While this is actually the fastest mechanism to provide support - you have to ensure that the version of E1 you're currently on will actually support the itanium platform. check minimum technical requirements.

Secondly, to echo Larry - itanium really sucks, its far too niche, and to be honest it doesn't provide anywhere near the customer base of AMD/Intel x64 on Linux. Just a thought there.

[ QUOTE ]


2. Migrate all E1 servers to Windows Intel platform and migrate Oracle database over to HP UX Itanium platform.


[/ QUOTE ]
Why not migrate the database to Oracle on Windows as well ? Moving the database to another platform, but keeping it on Oracle, is not a big risk compared to moving the code from Unix to Windows. If you're comfortable with Windows as a platform, just migrate everything to that platform. If you want the lower cost of Intel hardware but the reliability of Unix, then go with Linux. To be honest, Windows is just as reliable and scalable as linux is these days.

[ QUOTE ]


1. If we choose to migrate to Windows Intel platform for E1 servers (excluding database server), how much consultant work is required given that base infrastructre is in place?


[/ QUOTE ]
Moving the E1 applications from Unix to Windows isn't a big deal - in effect, you add the windows servers into your current environment, make sure your TNS Names and E1 Datasources are all configured and map everything through the OCM, then create a full package build and deploy ! It really isn't a big deal (obviously theres a few more steps, but thats the high-level view). Time to implement an E1 Application Server ? Start to finish can be accomplished in, say, 20 hours per application server (without tuning).

However, adding a new application server isn't a big issue - its the 3rd party applications that might be interfaced through that application server. Be careful of FTP applications etc.

[ QUOTE ]

We will go through vendor sizing exercise, but wondering how many Windows E1 servers do you think we need to handle 150+ concurrent keeping in mind that an upgrade to v9.0 will follow platform migration which is 100% thin client solution ?


[/ QUOTE ]
If you're running web clients and/or are mapping business functions to run on the application server, then I'd consider at least 2 application servers (quad processor, 4-8Gb of memory, fast drives - about two $10,000 boxes will be ample).

The UBE load is a big unknown. How many UBE's are processed each day ? What functionality is being processed ? Tell me what HP-UX box you're currently running, and the CPU load and the number of UBE's and I'll be able to tell you if you need any more CPU power for your batch processing.

[ QUOTE ]

3. At present we use HP Service Guard for failover for E1 servers, migration to Windows platform require re-visiting failover solution. We may need some loadbalancing solution as well as JDE next release is 100% thin client. What would be the best solution for Failover/Loadbalance.


[/ QUOTE ]
This is a big question. Are you referring to database failover or application failover ? Are you running citrix/fat clients or Web ? OAS or WAS ? How much uptime do you require - 24/7/365 or ca you have regular maintenance ? What about timezones ? How much of a high-availability solution do you THINK you need ? There are a lot of unknowns in this - and I'd need to audit your current architecture to provide recommendations on your future architecture.

Hope that helps. Remember, a Technical Audit is a great starting point for changing architecture and making recommendations for upgrade proposals. I charge a fixed fee of $10k for technical audits - and the recommendations, including accurate hardware sizing, can be in your companies hands within 2 weeks !

Hope that helps....
 
[ QUOTE ]
Fourth - where you really need to spend your hardware bucks is not on the servers but on your secondary storage. A really good SAN with a RAID 1+0 / 10 configuration will do far more for overall performance than a high powered server using a basic RAID-5 setup.

[/ QUOTE ]

Just to echo Larry's comments here. Data storage is absolutely where the performance really comes from. If you have high latency - then you really cannot perform correctly. Understanding the latest disk array technologies and how they affect E1 is paramount - and this is where an E1 Implementation Specialist with Storage Architecture experience really comes into play. Spend a few $$$ now on some good consulting experience, and you'll end up with an excellent designed architecture that you didn't overspend based on the hardware vendors recommendations !
 
Thanks a lot for all these great advises. I appreciate your input.

As advised, we probably go with technical audit for accurate server sizing and architecture. I agree that It's worth to spent $10k on a technical audit rather than overspending based on vendor recommendation.

Jon, here is the answer to some of your questions;

How many UBE's are processed each day?

I am not sure how many UBE's we process each day, but I guess we process around 10000+ jobs a day in production environment. Majority of these jobs are Purchase Order, Inbound Sales Order, GL Posting, Pick lists etc...

What HP-UX box you're currently running, and the CPU load?

We are running vpars on HP rp7420's servers. Batch server is currently allocated with 6 CPUs, but we never went over 15% of CPU usage.

Are you referring to database failover or application failover?

The reference here is to database and application failover. We use Service Guard for both database and application server.

Are you running citrix/fat clients or Web ?
80% of our users are fat client and 20% on thin. There is no Citrix user.

OAS or WAS?
We just finished with migrating to OAS. We have 2 OAS hosts with 2x OC4J with 2 JVM's on each OC4J each server.. WAS is still active to support Portal.

How much uptime do you require - 24/7/365 or ca you have regular maintenance?

We are not 24/7/365. We do have some maintenance Window on Weekends with only one Time Zone. We need a 100% high available solution except major disaster.

I understand all these questions are part of technical audit, but if you can shed some light that would be wonderful.

Thanks,
Sanjiv
 
Back
Top