Re: Informix Data Warehouses

jparker@epsilon.com wrote: > > Agreed - XPS kicks butt. But treat it like a DW, not an OLTP engine. That's exactly right, and why I mentioned, "in the context of a DW". We are getting guidance for DW that does not apply in the OLTP world, nor should it. The engine is quite robust, and I'm extremely happy with the shared-nothing design. We had one of the nodes go down for whatever reason, and the others kept humming along. I like it. And I've actually seen it work, which helps to prove it. > I wouldn't choose XPS to do OLTP. That means when designing your DW take > advantage of what XPS does extremely well - reading and writing massive > quantities of data, avoid working on a row by row basis with the data, > unless your doing sequential access. > Agreed. Get OnLine for OLTP. XPS has such a completely different orientation to DSS that I have heard will make it difficult for it to ever really be a part of the OLTP engine, or environment. And this is really a good thing, having a little specialization and focus. As to row-by-row, well this is where Kimball talks about the granularity of things, and despite the "summary" view DW tends to take on, there are still row-by-row needs and issues to contend with. Again, read the book. :-) And Jack I know you're on the ball about this, I'm just saying as a practical matter for everyone, read the book for the details. Inmon tends to diverge from Kimball so don't expect the same from these two camps. The subtlties can be profound. > Interestingly enough, pound for pound I have found HPL to be a faster > loader. with XPS I've only been able to hit about 1.93 GB/CPU/hr vs 2.3 > GB/CPU/hr for HPL. Of course an XPS platform typically has two or more > nodes and dozens of processors, so it's not a fair comparison. > > cheers > j. > I've not had an opportunity to test/compare, on a single-node system for quite some time, but we did get a copy of 7.23.? along with the XPS CD, so we do have the potential to try out the differences. I think the secret to XPS's success is in its simplicity. The real issues though for me are more hardware-vendor related than XPS, as IBM has turned their license manager into the extreme. We had to get cc licensed, only to find out Make wasn't a part of that. Geez, so then we got Make licensed, but it's ridiculous. According to our Informix ATG guy, we can take advantage of in-engine parallelisms even with ESQLC programs, using the standard cc. My impression from the past was that we'd need a pmake, etc. just to take advantage of the 4 nodes, but since the engine does the work, and it's talking through the switch, voila parallelism. So, it's a cool product, and smokes the living daylights out of Oracle. I might add though that fragmentation strategies are extremely important, and can be profound if not implemented correctly. This of course is true for XPS or OnLine. Apparently our site went through quite a rigorous evaluation of all the DW software and data base engines, and Informix was the ONLY one with a total end-to-end solution. Metacube has been optimized now expressly for Informix, and does not go to third-party engines anymore, which, on the surface was a big risk for Informix to do this, but it solidifies in my mind that Informix has made a serious commitment to classic data warehousing. And that is phenomenal. I think they really are going in the direction of being the best data base engine once again. There appears to also be good support in the market such as Business Objects with a Metacube tie-in product that works to produce reports, etc. Things are looking up. Now all we need is some marketing to go with it. Tim -- Tim Schaefer tschaefe@mindspring.com timohtim@netscape.net tim_schaefer@fpl.com http://www.inxutil.com Informix on Linux This Changes Everything