IAAI Version 0.2 pre-release

["Joe Pluta" <joepluta@xxxxxxxxxxxxxxxxx>]

XPOST: IGNITE/400, MIDRANGE-L, RPG400-L

This is just a quick message to pre-announce release 0.2 of the IAAI
Benchmark tests.  Currently, the benchmark suite is up to 42 tests,
comparing performance in both database I/O (native, embedded SQL, JDBC
and record-level access (RLA) through the JTOpen toolkit) and basic
decimal arithmetic (Java BigDecimal vs. RPG packed decimal).

The framework is robust enough to support a wide variety of additional
tests, and everything runs from green-screen commands (it took a bit of
magic to get that to work; thanks to Barbara Morris as always for her
boundless knowledge of things RPG, especially RPG to Java).  There is
still work to be done (load testing and multi-tier tests are not yet
supported), but I think you'll be impressed by what can be done.

Also, I'd like better reporting.  Right now I'm doing everything through
SQL and then pulling it into Excel.  I'd like to do better than that,
but that will wait for the next release.

I hope to put together the IAAI's first white paper this weekend.  The
results are quite conclusive, at least in my eyes.  Just out of my head,
I came up with a mix of operations that might represent adding a
"standard" order.  I figured 10 detail lines, and came up with the
following list of operations:

30 reads
11 writes
10 updates
30 comparisons
20 multiplies
20 adds

This included pricing and inventory adjustment.  Obviously this is a
very arbitrary number (and in fact probably pretty low), but I thought
it was reasonable for a first pass.  I haven't yet benchmarked
comparisons, but after adding in everything else, I came up with some
interesting numbers.

On my model 270, I can process 213 orders a second using RPG and native
DB2.  I can process roughly 23 a second with SQL: that's an order of
magnitude less performance.  Interestingly enough, RLA came in at about
the same pace, allowing 20 orders per second.  The big loser?  JDBC,
which only supports about 4 (FOUR) orders per second.

Note that this is without the overhead of connections in JDBC or file
opens in RPG; all of that is factored out in the tests.  But basically,
RPG and native DB2 run ten times faster than SQL and FIFTY TIMES FASTER
than JDBC.

All this should be available on the IAAI website Monday.

Joe

http://forums.plutabrothers.com/IAAI