Heavy Kaffe I/O at startup

Victor Zandy zandy at cs.wisc.edu
Thu Oct 15 16:01:25 PDT 1998


    First of all, Kaffe is very cool.  I'm grateful to all involved.

    Kaffe 1.0b2 generates a lot of very small reads when it starts on
my platform, Sparc Solaris 2.5.1.  I have counted nearly 10K calls to
read, most between 2 and 4 bytes, when Klasses.jar is read during
startup.  Some sequences of 20 of these, maybe longer, are contiguous.

    Does anyone know what this happens?  Why not read the .jar all at
once once?  Why not mmap it?  Why not at least read contiguous blocks
in one read?

    I couldn't find a way to search the the archive of this list, so I
don't know if this has been discussed previously.  If it has, a
pointer to the relevant messages would be great too.

Vic Zandy

More information about the kaffe mailing list