JRockit

~ the last words ~

Choosing a Java virtual machine was the kind of decision made by people who ran servers for a living, and i was never one of them. You get whatever the JDK installs and you go back to work. JRockit belonged to the other conversation, the one held by operations teams with a support contract and a load graph, and its whole life happened at that altitude.

It came out of Appeal Virtual Machines, a Swedish company started in 1998 by students from the Royal Institute of Technology in Stockholm. The first public JRockit landed in 2000 and made its name immediately on multithreaded server work. Version 2.0 for Solaris followed in march 2001 and version 3.0 that november, shipped with a management console that let you watch the heap of a running system, which in 2001 was not a normal thing to be able to do.

The technical position was blunt. Where Sun's HotSpot began by interpreting bytecode and compiled the parts that turned out to be hot, JRockit had no interpreter at all and compiled everything on the way in. That is a bad trade for a desktop application, where startup time is the thing a person notices, and a good one for a process that will be running for six weeks. Later versions added a garbage collector with an actual pause-time target rather than a hope, sold to people whose systems had to answer within a fixed number of milliseconds.

BEA Systems bought Appeal in february 2002, and JRockit became the JVM underneath WebLogic. That was a good place to be. For most of the decade WebLogic was the application server large Java shops standardized on, so a great deal of enterprise Java ran on a runtime built by a small Swedish team that most of those developers could not have named.

~ the record ~

~ what buying it twice was worth ~

Oracle bought BEA in 2008 and got JRockit with it. A year later Oracle bought Sun, and with Sun came HotSpot and Java itself. One company now owned the reference implementation and its fastest commercial rival, and there was never much doubt which one was going to survive that.

The announcement came at JavaOne in 2010: the two virtual machines would converge, with JRockit's distinctive parts reimplemented in HotSpot. In may 2011 JRockit was made free, which is what a product looks like when it has stopped being a product. The last public release, R28.2.3, shipped on 12 april 2012, and it only ever supported Java 6. After that it existed as a support obligation.

~ what came after ~

The convergence was not a polite way of saying deletion. JRockit Mission Control and its Flight Recorder, the tooling that made the JVM inspectable while it was under real load, were carried across into HotSpot and shipped with JDK 7 update 40 in 2013. They were open sourced in JDK 11. Anyone who has profiled a production Java service this decade has used a descendant of a Stockholm student project without being told.

Support for JRockit on Java SE 6 ran to december 2018, and Oracle's own release notes were still being revised in october of that year. So the runtime outlived its final public build by six and a half years, patched for customers who had paid to keep a dead JVM breathing under systems nobody wanted to touch. That is the most honest thing about this grave. Enterprise software does not get switched off. It gets billed until somebody stops paying.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »