Q. Web services: hype or real in addressing business integration issues?
A.
I think web services are a great idea but I haven't seen any great
multiple source integrated applications. Enterprise software companies
spent the last 5 years upgrading their platforms, all with independent
APIs if they have them at all, just to get onto the major J2EE servers.
But nobody hacks into their ERP anywhere near the web level. Also,
companies simply don't hire people at that level of skill (who can
understand the data model, fzample, and add other sources to it) to
build custom integrated apps.
I happen to think that there's something wrong with making the
browser the focus of all interactions on a desktop or laptop. If you
ask me, I think iTunes is the perfect example of an excellent client
server application. It's very well defined, it is extensible for third
parties to add value (like Last.fm) and it's beautiful. I like Turbo
Tax and I like Google Desktop too. It seems to me that as long as you
don't know exactly what you're doing, you build a thin client. When
your idea is baked you build a fat client. But already this Ajax stuff
is taking over the web paradigm.
I have a prejudice against web programmers. I think they pack too
much junk into their transactions and they really just build and build
without much forethought. I mean look how long it took SixApart to come
along after Vignette. So many CMSs are crap. I think it's because the
whole web services idea has way too many moving parts and the entire
paradigm was sustained because people wanted to get away from fat
clients.
So on the whole web programming and web services inch ahead very
slowly because the architecture is too wide open to force discipline on
its practitioners. Amazon and Google are really the two companies that
will end up defining web services, and it will all be through
rationally built APIs to their grid. When any and every programming
team can build collaboratively onto these platforms, then web services
will start to deliver.
Q. 1. What's the "best" search/database/data warehouse software/system, in
your opinion, that can handle larges amounts of data with ease and can
handle asking "questions" to which you don't know the question? In
other words, I have all of this sales data, but I don't know what the
data can really tell me outside of my pre-canned "questions".
A.
I'd probably overbuy in an enterprise scenario because I'd want all the
capacity I could handle so when I mention brand names, you can
certainly scale down from there. But there are a few components that
would be absolutely necessary.
First and foremost (oddly) I'd get a metadata management system that
has the ability to generate attributes based on dynamic business rules.
That way, given any set of data, I can reclassify it according to
things I find out about it later. My pick is the only one I'm familiar
with which is Razza/Hyperion MDM.
Secondly, I'd get an industrial strength database for standard
warehousing. I tend to go with DB2 because of its relative programmatic
inflexibility as compared to Oracle and MSSQL. I don't want to do
anything fancy in the relational layer and the proprietary extensions
in T-SQL and PL-SQL make programmers get creative. I actually believe a
lot of incredible things about Teradata and Sybase ASE, but I have no
experience with either. I'd house this on a generic 6 or 8 way box with
a huge SAN and use it to vacuum up all enterprise data for
multidimensional staging.
Thirdly, I'd definitely go with Hyperion Essbase Analytic Services.
The new Aggregate Storage Option on 64 bit can serve up thousands of 30
dimensional subsecond queries. Basically more than your brain can
consider, faster than you can think it for hundreds of people
concurrently. I'd actually spend more money on this interactive query
platform than for the DW backend. I'm partial to HP's 64Bit hardware.
Finally, I'd use regular Excel as one of the front-ends and Tableau
for the other. Anybody who is not capable of handling the complexities
of reading multidimensional data in Excel is similarly incapable of
understanding the implications of the analysis to the business. I have
unsubscribed to the belief that computer systems will make things plain
to those who don't understand the business.
You have to have some idea of the relations between profitability
and cost of sales to put those two things into a two-way analysis, but
then to put in employee benefits and freight costs as well and think of
a management program that ties them together as befits your analysis is
the level of thinking necessary. With this kind of system in place, you
can rapidly build query spaces for that kind of analysis *and planning*
and change your budgets and forecasts appropriately. By rapidly I mean
in one day.
Given staffing and competence in the technologies it would take one
day from transaction system to denormalized ODS to n-dimensional model
to front-end. Both of these front-ends to Essbase are zero
configuration, instant query mechanisms. which dynamically understand
changes to the underlying master data. With MDM in place you can take
results of your queries feed them back and restage the data every day.
This is what I call a visual human-based data-mining paradigm. It
presumes that the dimensionality of the incoming data is somewhere less
than about 20, but its extensible.
A. I think web services are a great idea but I haven't seen any great multiple source integrated applications. Enterprise software companies spent the last 5 years upgrading their platforms, all with independent APIs if they have them at all, just to get onto the major J2EE servers. But nobody hacks into their ERP anywhere near the web level. Also, companies simply don't hire people at that level of skill (who can understand the data model, fzample, and add other sources to it) to build custom integrated apps.
I happen to think that there's something wrong with making the browser the focus of all interactions on a desktop or laptop. If you ask me, I think iTunes is the perfect example of an excellent client server application. It's very well defined, it is extensible for third parties to add value (like Last.fm) and it's beautiful. I like Turbo Tax and I like Google Desktop too. It seems to me that as long as you don't know exactly what you're doing, you build a thin client. When your idea is baked you build a fat client. But already this Ajax stuff is taking over the web paradigm.
I have a prejudice against web programmers. I think they pack too much junk into their transactions and they really just build and build without much forethought. I mean look how long it took SixApart to come along after Vignette. So many CMSs are crap. I think it's because the whole web services idea has way too many moving parts and the entire paradigm was sustained because people wanted to get away from fat clients.
So on the whole web programming and web services inch ahead very slowly because the architecture is too wide open to force discipline on its practitioners. Amazon and Google are really the two companies that will end up defining web services, and it will all be through rationally built APIs to their grid. When any and every programming team can build collaboratively onto these platforms, then web services will start to deliver.