Xe vs. 8.12 performance/size guesstimation question...

[swhitmire]

Ok, I realize what I'm about to ask has no right answer, and could all vary drastically by platform/type of workload/size of database/CNC's earlobe length. So what I'm looking for is not any attempt to guess what it would be like for my environment, but rather some data points to start my thought process from. So, from people who've upgraded from Xe to 8.12 (or 9, if anyone's gone so far), I'd like to get a vague guesstimation of two things: 1) what level of performance impact/memory usage increase on your enterprise servers you saw, if you did it on the same hardware, and 2) how much bigger your total database size is now (assuming you converted business data to Unicode). I'm guessing the answer to #2 is going to be 40%-60%, from previous discussions we've had here, but I'd like to see a few people's specific numbers, if anybody's got them. I've really got no grasp on what the answer to #1 is going to be... from past posts I've gathered that the impact has been more significant than people have expected in some cases, but nothing more specific.

 

[gregglarkin]

Scotti,

My earlobe length is pretty normal, but thanks for asking. the 40%-60% growth of the database is accurate because of the unicode conversion and the usual vendor bloat factor. we never convert to an exisiting machine, it's always new hardware, so I can't give you a ballpark figure on performance.

Gregg "nice earlobes" Larkin

 

[cmanderson]

As always, a professional review from your hardware vendor is a good idea before undertaking an upgrade of your ERP. If I had it to do over again, I would have replaced hardware when upgrading from Xe to 8.9, but it was still under lease, depreciation scheduled count sometimes, etc.

On an HP-UX based machine, specifically an rp8400 with twelve 875 MHz CPU's, I recall a roughly 20-25% increase in overall CPU utilization when we moved from Xe to 8.9 in 2004. This was with my former employer. In hindsight a migration to Itanium would have been premature, but a migration to AIX on newer P4/P5 hardware would have been ideal. We also added two more CPU's to that box to bring the total to 14, prior to migrating batch processing from HP9000 hardware to IBM System p with AIX in 2006 or 2007.

To make my data points semi-relevant, I would add roughly, 2-3% of memory utilization for each major Tools Release to have shipped since SP2, which is what we ultimately went live with for 8.9. 8.93 was soon to follow, then 8.94, jumping to 8.96 and ultimately 8.97 was where they were at prior to their upgrade from 8.9 to 8.12 (after I left.) After a couple of major upgrades (8.94 and 8.96, specifically), I can recall UNIX system admins and DBAs noticing overall utilization increases for the same hardware, all other things remaining equal.

Keep in mind your PDF output in the PrintQueue location will roughly double in size due to the conversion of system code to Unicode. This is true with our without a Unicode conversion of the Business Data/Control Tables. For some shops this might not have a big impact, for others it could be very expensive.

You could also see (roughly) a doubling of memory utilization for certain jdenet_k kernel processes on an AIX server, for instance. Especially the call object kernels.