Along the way we evaluated Sun, HP, and DEC hardware (NB was adamant that they were not interested in IBM systems). We mostly looked at various UNIX variants (System V, BSD, and OSF-1), although DEC kept saying things like "Yes, yes we have Alpha based systems running OSF and System V. You really should be looking at VAX/VMS systems!" But NB was convinced that proprietary OS's were the past, they'd just been bitten by Prime/Primos, and wanted to go with a UNIX variant. We had meetings with Oracle, Sybase, Informix, Unify, Progress, and Ingres including some demos and discussions with developers and support engineers about features and internals.
Anyway, we carefully gathered specifications, counted lines of code and numbers of modules, evaluated plans for new projects in the pipeline that had been halted after Prime's announcement, met with hardware and database vendors and after maybe six months or so we made our recommendations. Here they are with a concise brief of our reasons:
- Hardware - Sun server blades
HP and DEC each had multiple OS's and hardware platforms. On the OS side, UNIX wasn't the main product for either company. We were concerned about their commitment to UNIX in the future and Sun was 100% UNIX and was gaining ground in the industry.
- OS - UNIX System V - Solaris
- Database - Informix OnLine v4.01
NB didn't like the locking model on some of the databases. Oracle didn't have any development tools beyond a very simplistic form manager and an ESQL implementation that was incomplete against the developing standard. Also Oracle's SQL dialect did not conform well to the SQL-89 standard. Informix OnLine had row level locking, stored procedures, triggers, referential integrity, and was the first RDBMS that could be archived successfully while fully online (hence the name) processing transactions. Informix had started as a development tool company that happened to have a great database, so there was a very complete ESQL/C implementation as well as a 4th Generation Language (4GL) and high level menu, forms, and reporting tools for building quick applications when responsiveness to business requirements was needed. Sybase was similar in design to Informix, but, at the time, it did not have an ESQL/C implementation at all, only its Open Client (OC) library. While several years later OC became the ODBC v1 interface library standard, at the time it was purely proprietary to Sybase.
- Development environment - ESQL/C (for its portability across SQL databases)
They said "Thank you." for our recommendation report, hired both of us full time, and then went ahead and bought:
- Hardware - DEC Alpha - because DEC's stock was performing better in the market than Sun's (NB was, after all, a financial organization).
- OS - OSF-1 - because both DEC and IBM were driving the development - "How can it not be big!"
- Database - Sybase - because all of Wall Street was using Sybase, so NB should also! Oh, and Sybase promised to deliver an working ESQL/C in 90 days.
Anyway, long story long, we spend four months getting things up and running on the new systems and ported the first application, NB's client billings and invoice generation. First problem we ran into was how to represent portfolio values. Some financial instruments were priced in Japanese Yen, Italian Lire, and Hungarian Lira. Any client that owned many shares of any of these securities would have either a native currency value with more whole unit digits or a US currency balance with more fractional digits than a Sybase DECIMAL type could hold (15 total digits fixed as 12 to the left of the decimal and three to the right). The head of development and I attended a Sybase Financial Users Group meeting and asked them "What are you doing about this? Do any of you have, or know about, an Open Server module that can handle larger numbers?" (Open Server was a library of functions one could use to add functionality to Sybase - a unique feature at the time.) The answer was basically that "there are no such Open Server modules" because no one was willing to use the Sybase Open Server libraries to basically build a customized server (SOS modules were linked into the base server). "Then how are you handling the larger numbers for big customers with Yen and Lira securities?" we asked. The answer was "We don't have any such, and we're hoping we never do." NB did. (For what it's worth, Informix's DECIMAL type supported up to 32 decimal digits of precision, either floating point or configurable fixed point.) Strike one!
Then we started to really code the application and found that the ESQL/C that Sybase delivered was a) not ready for prime time (apps built with it kept crashing) and b) was brain dead, and missing several features of the language that we needed like named cursors and multiple cursors in a single application. Sybase wanted 12 months to get those features implemented for us. Strike two.
Strike three was performance. Our initial tests were that our brand spanking new DEC Alpha, cutting edge systems, running Sybase, would take up to 4 days to process the data and generate the invoices that the several years old, out-of-date technology of the Prime systems could get done in around one day. Strike 3.
We called Informix, loaded up Informix Online v5.02 and ESQL/C, exported and imported the database (again), recompiled the code improving it using the features we couldn't get from the Sybase ESQL/C, and wham! We were done with Billing and invoicing in another month or so and getting the invoice runs done on a Sunday afternoon (OK after they finally went out and bought the faster and higher volume printers we had recommended because we were now generating reports faster than the operators could change the paper in the printers they originally purchased - forgot that part). Later other interactive systems were written in Informix 4GL and even a couple using Informix's Forms, Menus, and Reports modules. Go Informix!
Let me know if you want to hear more stories like this.