The test measured how many times HP 3000 languages could execute the Sieve of Eratosthenes, to locate the first 16,384 prime numbers, in 10 seconds. Java/iX, running on a Series 988, showed a best-case HotSpot compiled performance of 288 times, while C’s performance ranged from 158 to 702 times. Optimized COBOL managed the Sieve task 182 times.

Yawn presented the test results at this year’s SIG Java meeting at HP World, and was quick to point out that benchmarks can be tailored to make anything look fast or slow. But the one-page of benchmark code for interpreted languages like the MPE Posix shell and Perl (slower than Java/iX), and compiled languages like COBOL (slightly behind Java/iX), were hand-tuned by CSY lab staff, he added.

“If you look at the whole range of compilers, Java comes off a lot better than if you compare it to optimized C,” Yawn said. “But not everybody’s running optimized C. A lot of people are doing things in command files. All interpreted languages had scores of 2 or less. If you’re doing things in an interpreted language, you would probably see better performance going to Java.”

“We’re in the range of compiled language performance,” Yawn said. “If you try to conclude anything from this, your conclusion would almost certainly be wrong. But it said to me that running Java would compare very favorably with interpreted languages, and we look very competitive with compiled languages on the HP e3000.”

The HotSpot Java compiler, due out for the 3000 in the MPE/iX 7.0 Express 1 release and now in beta test, “puts us competitive with COBOL,” Yawn said. “That’s where we really need to be on the platform.” It speeds up memory management and garbage collection, he added.

The benchmark does no object creation, a typical Java task that affects performance. It has no IO like most HP 3000 applications, but Yawn said that CSY may “get around to releasing some numbers for a TPC benchmark written in Java.” He added in a quip at the Java meeting, “We might do it, but only if we like the answers.”

All four variations of Java tested are based on the Java 1.2.2 JDK kit, available as a supported download from the CSY Jazz Web site. And that HotSpot release of Java is coming. “We’ve got a road map of doing performance enhancements more aggressively than most of these other languages will be doing performance enhancements,” Yawn said.

Understanding the language is the key to getting top performance out of Java, he added, a rule that holds for any compiler. “If you’re a very experienced COBOL programmer and a really rookie Java programmer, your first programs probably won’t run as fast as your COBOL programs,” Yawn said. “There’s lot of Java profiling options to help you get information and move your performance along.”

Servlets and app servers

Another way to boost Java performance involves doing less startup overhead for applications on the 3000. Servlets and application servers make this possible, incurring the startup penalty by eliminating the process creation associated with CGI creation. “Servlets is a way to do Java code that runs inside the Web server, compared to applets, which run in the browser client side,” Yawn said. Servlets allow for session persistence and have faster startup times than traditional CGI once the first user in has started the 3000’s Java Virtual Machine.

The HP 3000 has beta-test versions of servlets for the 1.3.9 Apache Web browser, both in its secure and non-secure versions. The software, supported through e-mail only while it’s in beta, is available from jazz.external.hp.com/src/jserv. Sites must be running MPE/iX 6.0 or later, Apache/iX 1.3.9 beta or WebWise Secure Web Server for MPE/iX; install beta test patch MPELX44 for 6.0 or 6.5 (which fixes a DSO-related problem, and requires an HP support contract) and any of the Java/iX versions: 1.1.7B, 1.2, or 1.2.2.

The free Enhydra application server, running on HP 3000s in production (see related story, page 1) and Java servlets are similar, “in that they reduce the startup overhead by running continuously,” said HP Response Center engineer Lars Appel. “In fact the Enhydra application framework is a Java servlet itself.”

Yawn said that Web servers by themselves don’t offer the advantages of an application server like Enhydra, which provides transactional access to data, security, resource pooling and failovers. “If you had a Web server only, you’d have to write all those pieces yourself,” he said.

Creating Java applications using an application server “is simpler,” Yawn noted in a recent chat on 3kworld.com, “because many of the things you might otherwise have to worry about (like multithreading) are done for you by the servlet engine or application server. So you just worry about your business logic.”

“In the last year it seems like Java has gotten a lot more real for a lot of people,” said SIG Java co-chair Gavin Scott of Allegro Consultants. “A lot of people are starting to roll out applications.” Scott recommended Bruce Eckel’s “Thinking in Java” as the best single book to learn the language. The book can be downloaded free off the Web as a 4Mb PDF file from www.planetpdf.com/codecuts/pdfs/bruceeckel/TIJ2.pdf.