MDA and MDD

Looks like we are moving down the Model Driven Architecture and Model Driven Development road. Yesterday everyone on my development team had a quick training session on Borland Together. We are using the Eclipse edition. The reverse engineering features of the product are very cool, although I'm not convinced yet about the Pattern Recognizer. The Quality Assurance metrics and audits look interesting - hopefully you will be able to run these from the command line, and generate them via the nightly Ant build.

Database Usage

This is an old blog entry by Werner, but is still relevant today. I'd be curious to know if anyone has any information on how these sites have evolved since 2003.

Gigaspaces and GemFire

Gigaspaces and GemFire both appear to be off-the-shelf products that can offer caching (distributed). Both products implement JCache (JSR 107). GemFire offers the option of using an operating systems shared memory segment to improvement performance between processes on the same machine. Has anyone used GemFire? Any idea which is the better product?

C++ Migration to Java

After 3 months of work, I'm close to being done with the re-architecture of three legacy C++ servers. The original servers were written back in the late 90's, and used an OO database as a repository. The re-architecture has involved moving and code to Java, and migrating off the OO database. What follows is a summary of the performance timings taken from a JUnit calling both the C++ and Java CORBA servers - these timings should be taken with a pinch of salt, since the internals of the C++ and Java servers are different (but produce the same results, and expose the same IDL interfaces), as is the repositories:

(11998 method calls) for Test 1
C++:Average milliseconds 21(total milliseconds:253846)
Java:Average milliseconds 1(total milliseconds:18438)

(11998 method calls) for Test 2
C++:Average milliseconds 26(total milliseconds:314209)
Java:Average milliseconds 3(total milliseconds:38745)

(5999 method calls) for Test 3
C++:Average milliseconds 40(total milliseconds:242797)
Java:Average milliseconds 1(total milliseconds:10684)

(5999 method calls) for Test 4
C++:Average milliseconds 53(total milliseconds:322152)
Java:Average milliseconds 11(total milliseconds:69485)

(10 method calls) for Test 5
C++:Average milliseconds 717(total milliseconds:7174)
Java:Average milliseconds 168(total milliseconds:1680)

(10 method calls) for Test 6
C++:Average milliseconds 101(total milliseconds:1019)
Java:Average milliseconds 25(total milliseconds:259)

(10 method calls) for Test 7
C++:Average milliseconds 81(total milliseconds:813)
Java:Average milliseconds 28(total milliseconds:287)

(813 method calls) for Test 8
C++:Average milliseconds 3(total milliseconds:3133)
Java:Average milliseconds 2(total milliseconds:2053)

(813 method calls) for Test 9
C++:Average milliseconds 16(total milliseconds:13629)
Java:Average milliseconds 1(total milliseconds:1237)

Timings were taken on 8 CPU (1200Mhz) Sun Sparc Solaris box running JDK 1.4.2_05 with Iona Orbix 6.1

Q: Why are there two ps utilities on my Sun box?

Answer
Historically, SunOS 4.x was based on BSD unix. Solaris 2.x (= SunOS 5.x) was based on SYSV, with a bunch of commands having different syntax and behavior. To ease the transition, the/usr/ucb directory was created to hold the incompatible BSD versions. People who really wanted BSD could put /usr/ucb before /usr in theirPATH.