By adding a twist of Linux to a simple database upgrade, Sean Lincolne's IT team mixed up a tasty ROI martini. Lincolne -- systems manager for the Australian Fisheries Management Authority (AFMA) in Canberra -- describes the testing, implementation and results of an OpenIngres 1.2 and Linux migration, as he continues the story he began in part one of this interview. He also discusses future Linux migration plans. In part one, he described...
why NT was the wrong choice for AFMA's DBMS.
You've described how the Linux decision was made. Now, can you describe the OpenIngres-on-Linux planning and design process?
Sean Lincolne: The beauty of back-end systems is that, as long as they deliver the required service to end users, the [users] largely don't care how you do it. This upgrade was an internal project for the systems section and so did not require any extensive consultation with the organization at large.
The initial planning and design process largely involved us convincing ourselves that we could obtain the necessary services on the [Open]Ingres-Linux servers as the existing Ingres-HP servers were providing, but at an equivalent or better standard.
Discussions with Computer Associates were quite positive, and we had time to try it out, so we decided to 'trial' it.
What happened in the trials?
Lincolne: We built up a test server using a desktop computer, replacing the IDE disk with a pair of SCSI drives and an Adaptec Ultra-320 SCSI controller. Then we installed Red Hat Linux 6.0 and the beta release of [Open]Ingres, copied some of our larger, more complex business systems across, and tested.
The testing process involved recompiling all the required applications using the Ingres tools and GNU GCC compiler. The Ingres applications compiled flawlessly, but the GCC compiler was a surprise. With all warnings turned on, GCC flagged some problems with the code we were migrating and gave us the option to raise the quality of some of our internal systems. This was viewed quite favorably by our application developers.
After testing, what were your conclusions?
Lincolne: The testing was quite an eye-opener. The performance of the system was excellent, and the ease of migration was impressive. Based on the trial, it appeared possible to:
-- Replace the main HP systems with new, commodity hardware;
-- Acquire additional equipment to provide on-site hot spares; and
-- Eliminate the need for specialist support contracts and subsequently save money.
We were also able to use the Linux 'pam_smb' module to replace the Ingres authentication code with our own internally developed code. This allowed us to provide DBMS account authentication against our existing Windows NT domain, eliminating the need for users to maintain yet another password, with a subsequent reduction in support requirements.
In short, it looked like one of those offers you just can't refuse.
How did the deployment go? Were there any glitches, unpleasant surprises or pleasant surprises?
Lincolne: Deployment went like clockwork. It was simply a case of bringing the databases on both old and new systems in sync, stopping the DBMS on the old system, and reconfiguring applications to connect to the new server.
As we do not run a 24/7 operation, we could safely schedule work outside core business hours. If there were any problems, we could switch back to the old servers while we sorted things out.
There were some users who used Telnet and needed some shortcuts and profiles updated, and a minor termcap issue that was quickly fixed, but beyond that, no problems.
What has been the impact of the new implementation on server costs?
Lincolne:By replacing the two HP D320 multi-processor servers, we saved AU$44,000 per annum in support costs. To put that in perspective, our current budget of approximately AU$1 million is roughly half operating expenses. So, the cross-grade has resulted in an ongoing savings of approximately 8% of our IT operating costs.
The replacement Dell desktop computers with additional SCSI disk and tape drives and three-year, next-business-day support cost approximately AU$12,000 for the pair. We also took the step of buying in some spare disk and tape drives to ensure minimal downtime should something fail.
What was the impact in the areas of software and productivity?
Lincolne:The Ingres DBMS upgrade was scheduled to occur, so any additional staff time involved was minimal, and probably offset by allowing us to upgrade while retaining the old HP servers as a fallback in case there were problems. The DBMS license cross-grade cost us a couple of phone calls. Computer Associates were very helpful there.
Testing took several days for two staff, with some involvement from end users to confirm that all was well.
My feeling is that we managed this implementation and recouped our costs within the first three to five months, while providing improved service to end users, a reduced administration maintenance load on systems staff, and a substantial saving in ongoing operating costs.
Speaking of users, what did they think of the change?
Lincolne: Things were faster for end users. Beyond that nobody noticed a thing.
What could have turned out better?
Lincolne:Not a lot. At the end of the day, our expectations were exceeded.
My only regret is, again, the lack of support from other application vendors. I'd love to replace the last of our HP Unix servers with Linux, but the application vendor support is missing.
So, what does AFMA's data center look like today?
Lincolne:The back-end systems consist of a mix of Windows NT servers, Linux and HP-UX minicomputers. The NT servers host 'standard' tools such as file and print sharing, as well as the Exchange mail server.
Our network environment is largely a 100BaseT switch topology, with a scattering of older 10BaseT legacy devices (printers, etc.).
Do you have plans for more Linux migrations?
Lincolne:In the next two to three years, we will be due to replace some of the other back-end systems, specifically file and print services, and Samba is looking like a possibility.
Our two remaining HP A180C servers currently support our payroll and finance systems. Given that these servers literally cost more than they are worth each year in maintenance costs alone, I am looking forward to retiring them as soon as feasible.
We are in the process of developing and testing our migration strategy to move the finance system to Oracle on Linux, and progress is promising. I expect that the production implementation will occur some time [in the] next financial year.
The payroll system is currently not supported on Linux. Hopefully, this will change in the future.
We are currently piloting some e-commerce applications using S/MIME and XML, with external independent software vendors developing front-end systems for end users. The main idea is to ensure that we get the data we need, without unnecessarily constraining external clients' business processes. It's an interesting project and, needless to say, Linux has a role to play here.
On the two-to-three year horizon, there are several NT servers due for replacement and, again, Linux may have a role to play there.
FEEDBACK: Is your enterprise moving off NT to Linux? We'd like to tell your story.
Send your feedback to the SearchEnterpriseLinux.com news team.