Back to the Real World and RAC Architecture

I find it amazing that my last post was nine days ago already. It really seems like yesterday that I was still in San Francisco for OpenWorld. To close out that thought, the Aerosmith concert was pretty good, but not great. The could have played a little longer and done some more of the older stuff... (I was really hoping to hear "Big 10 Inch"). It is pretty amazing that they are able to put on that kind of a show at that age. I can only hope to be that active.

Thursday was actually not too bad mostly because I knew I had to be up early so I took it easier at the party than in past years. I was invited to the Inner Circle event for Oracle Fusion Applications because of my involvement on the Board of Directors with IOUG. This was my first event like this and it was very interesting. Most of what we saw and heard was not new to many in the audience because they had been taking part in the testing, but to me it was mostly new. The development and testing cycle they have been using was the most interesting part to me. It seems that Oracle is taking customer input very seriously. They have had users looking at the applications all along the way and taking the feedback and putting it into the applications. Hopefully that makes a better product for all of us.

For my company, the best part I heard was the ability to run Fusion apps along side existing applications. We are looking at adding some new modules to PeopleSoft and also bringing in some project management software. From what they showed us, the Fusion applications are built so we could bring in HR and Primavera on Fusion and seamlessly integrate with our existing PeopleSoft Financials system until we are ready to move that over. This sounds pretty nice and makes the switch seem easier than doing a full "fork lift" migration. I closed the trip with a PeopleSoft on RAC presentation and a late night trip home. All in all OpenWorld 09 was a great experience. Reconnecting with many people and putting faces to the phone voices make the trip worthwhile. Adding the content and the new people to be met are icing on the cake!

So, back to the real world. Part of the reason for attending the PeopleSoft on RAC session is that we are currently looking at replacing/upgrading our database infrastructure for PeopleSoft. I have spent part of the last couple of days researching and pricing a few different options for RAC. We currently have a shared HP Rx6600 that has eight cores and is virtualized. The database only has access to four cores and we are almost at capacity. The original plan was to purchase another Rx6600 and continue on that platform, but we are unsure of our growth over the next few years and think RAC would be a better option to give us the ability to scale as needed. That and the fact that the Rx6600 is Itanium are leading us to look at Intel based servers running Linux.

My main research points now are whether to go with more sockets with fewer cores or fewer sockets with more cores. I can see the pros and cons to both and will probably end up with fewer sockets with more cores, again because it gives us more flexibility for growth in the future. The other question is the interconnect. I'm sure gigabit ethernet would be okay, but I want to make sure there are no issues moving to a brand new platform. I'm including Infiniband switches into my design and figure I can always look at something less if we need to cut costs.

I have had to architect new solutions in the past and I always feel a little like I am putting my career on the line. That is where good research and good professional networks come into play. As long as I can back up my recommendations with multiple sources I am comfortable that things will work out. The user groups help here quite a bit (you didn't actually think I'd finish this without a user group plug). I have contacted three different people over the last couple of days for advice on my architecture. All have given me good advice and other sources of relevant information. I still haven't submitted an abstract for Collaborate 10 yet and they are due by midnight. Maybe this architecture research would be a good option...