[lug] OT: Enterprise DB Migration

[Glenn Ashton]

Kyle,

I have migrated legacy databases before and have used Oracle, SQL Server
7.0 and SQL 2000.

SQL 2000 has an impressive list of features and is priced right compared
to Oracle.  That having been said, there is not alot of experience other
than Microsoft's propaganda (and their partners).  According to an article
I read recently by Neil Gunther, a capacity planning consultant, one of
the nettlesome issues in comparing Enterprise Database solutions is that
with Microsoft products vs. Unix based products you are often comparing
apples to oranges especially considering the underlying hardware
architectures these OSes can be run on.  This is particularly important in
understanding clustering, scalability and parallelism with respect to
implementation of production database systems.  (Gunther is the gentleman
who derived the Gunther curve, which demonstrates the limitations of
assuming SMP is the magic bullet for scalability, BTW)

See the following article:
http://www.teamquest.com/html/gunther/gunther.shtml

There are also some good articles on the performance of NT systems and the
use(or misuse) of the Transactions Processing Council's benchmarks at his
company's website: http://www.perfdynamics.com/

In basic terms, the requirements that you have to analyze include the
following:

1)  Do I have to impliment parallelism in the architecture?

2)  How many rows of data? 100,000? 1,000,000? 10,000,000? 5?

3)  OLTP? Warehouse? Data Mart? ODS?

4)  Cluster/ High Availibility Requirements?

etc. etc.

In the case of SQL Server 2k, they promise parallelism in the software,
but you actually have to buy a packaged hardware and software solution to
make it go, for example.

These kinds of debates can get very religious very quickly, my take is
that any selection has to be based on hard requirements and not marketing
fluff.  The other consternation is that both Oracle and MIcrosoft won't
let you publish benchmarks of their products- violates their license
agreement.

I think it is very important that your search be based on hard,
quantifiable performance requirements and that you don't get caught up in
fluff.  Most of the vendors will allow you to acquire evaluation copies of
their databases to play with.  If time and machine resources are
available, this can be a good thing, so long as you remember to compare
apples to apples and oranges to oranges (unless you find a lemon).

Also, understanding George's comments regarding the other things that you
get roped into when you make a Microsoft decision is very important.  I
think that with regard to a web based application port, there are other
significant questions regarding web and app servers etc.

Please feel free to email me off list and I'd be happy to give you a call.

Now, if only there was a real Enterprise class transactional database that
was free and Open Source... (and no, PostgreSQL doesn't count...)

-Glenn Ashton
gfa at idiom.com