Re: Should I replace all CHAIN, SETLL, and READs to SQL

[Nathan Andelin <nandelin@xxxxxxxxx>]

> But in return for that 100 millionth of seconds of performance you are

> hardwiring the physical view of the table into the program.

I don't have a problem hard-wiring external data structures into programs. That concern was already addressed by others in the other thread. We have a CL command / program that reports which programs / service programs may be affected by changes to tables, and optionally compiles them automatically. Not a big deal.

Yes, improving I/O performance by a factor of 5-6 is material to me. I've run benchmarks where RLA was up to 25 times more efficient than SQL. It depends on what you're doing, and it's good to know what you're doing.

Forget about the I/O performance for a moment and consider the total resource costs of loading the SQE in every job launched. The SQE is a substantial (big) run-time environment in and of itself. You could code *JAVA string objects in all of your RPG programs too, but would you want to load a JVM for every job?

> You have lost your database independence for a few
> 100 millionth of second performance

You lose database independence by using RPG to begin with.

> I have said over and over the whole performance thing starts going
> out the window when you begin to consider multiple tables.

Repeating it over and over doesn't make it any more true. Whether you use SQL or RLA you should know what you're doing and understand the performance implications. Under the covers, the SQE uses RLA. Why should IBM be the only one to code an "engine".

-Nathan.