Also, "acceptable performance" depends on your application. If you
want to write big numerical codes that run for three months, stick with
FORTRAN. If your Java app spends most of it's time waiting for queries
to get back from a database, information to come over the net, or for
a user to rain keystrokes on it, than the difference between Linux with
(or without) TYA and the fancy commercial JIT's on Windows is miniscule.
Something I'm wondering about -- the blackdown people are pushing
the "Native Threads" version of JDK 1.1.7 as an exotic item for people
with SMP machines. What I've been noticing, however, is that
(a) Most of the other unix implementations use kernel threads now
(Solaris, AIX, Digital Unix),
(b) kernel thread semantics are different from user thread semantics,
(c) kernel thread semantics are more like Win32 thread semantics (than
user threads are)
(d) most java development and deployment is on Win32
(e) therefore, a lot of java applications are going to require (either
by design, or more likely by accident) kernel threads to work properly
Just a crazy guess on my part. But I've noticed that it usually
takes two hours to knock off an applet and two weekends to get the
threading to work right on Unix, Win32 and the Mac.
