sionlib Version 1.1p7: (September 2009)
---------------------------------------
--> see http://www.fz-juelich.de/jsc/sionlib/

  - New Features in 1.1p7:
      □ more function to get information about SION-file (e.g.
        sion_get_current_location, sion_get_bytes_written, sion_get_bytes_read)
      □ error handling enhanced
      □ Fortran interface enhanced
      □ siondump: new option -m (print map)
      □ partest: new option -Z offset, read data from other task to prevent
        cache effect in measurement
  - Updates/Bugfixes 1.1p7:
      □ fixes for performance improvements on IBM Blue Gene/P
      □ stabilized, more tests on different platforms
      □ partest: increased MAX_PE to 288k

sionlib Version 1.1p6: (May 5th 2009)
-------------------------------------
--> see http://www.fz-juelich.de/jsc/sionlib/

 - New Features in 1.1p6:
   - new wrapper functions sion_fwrite, sion_fread (beta)
   - automatic determination of file system block size, if parameter is -1


sionlib Version 1.1p5 (beta): (February 4th 2009)
-------------------------------------------------

 - first down-loadable version 1.1p5 of sionlib (beta)
   --> see http://www.fz-juelich.de/jsc/sionlib/

 - New Features in 1.1p5:
   - Multifile option (one logical sionfile consists of serveral phsical files) is 
     implemented and tested
   - Fortran Interface is implemented and tested
   - parallel benchmarking programs partest and fpartest for sionlib and MPI/IO available
   - sionlib with Fortran Interface is used in production code PEPC 
   - Doxygen documentation available
   - serial utilities: sionsplit, siondump and sionmerge 
   - examples programs for converting application specific data to sionlib files

 - Tests:
   - tests on Blue Gene/P up to 64k cores, 2 TB files --> 90% of Peak bandwidth
   - tests on IBM Power 6 and Linux-Systems
   - first functional tests on Cray XT4 


- Current open Issues:
  --------------------
   - more testing needed

   - Multi-File-Support in serial API (sion_open, sion_open_rank, sion_seek)
     -----------------------------------------------------------------------    
      Changes for including mapping from task to physical sion files (multi file support)
       -> added mapping (grank -> filename,lrank) to first sion file 
    - sion_open_mpi (done in 1.1p5)
      - (rd) reads mapping on first grank and distribute filenumber and lrank
      --> collective open without overhead, 2*gsize*4 Bytes add. space in file needed
    - sion_close_mpi (done in 1.1p5)
      - (wr) writes mapping, first rank collects and write it
    - sion_open_rank (done in 1.1p5) 
      - (wr) not implemented as proposed by design
      - (rd) each task has to open first file and read mapping entry for own grank,
            if necessary, sion_open_rank will close internally first file and open one 
            of the other files
        --> in average, each task has to open 2 files, but only to read a few bytes from 
            the first one 
    - sion_open (tbd) 
     - (wr) currently only one physical file implemented, multi-file support necessary?
     - (rd) sion_open opens first file and read all metadata in internal structure
            if current file is first of a multifile set:
             - it reads also the mapping table and stores it in internal structure
             - it allocates an additional vector with one integer element for each file
    - sion_seek (tbd) 
      - no changes for parallel access because rank will not be changed during operation
        on file
      - has to get one new parameter ( ..., FILE **fp) for support in serial access
      - has first to check if new rank is also in current file, 
        if yes move file pointer and return with current filepointer 
      - if new rank in other file
        - open new file if neccessary with sion_open (recursively) and store sid in 
          vector
        - calls sion_seek on this file with same parameters
        - returns file pointer which was given by the sion_seek before
       --> limitations: large number of open files possible and for each file complete 
                        meta data is stored in memory
       --> only initial overhead due to multiple open, but file pointer will be cached
           fast changes of rank and or position in file possible
    - Hints when using less number of task for reading than writing:
      - use multiple sion_open_rank on each task if the number of locations to open in 
        sion files is moderate, 
        e.g. (written with 512 task, read with 32 tasks --> 16 open files/task)
      - use one sion_open on each task if the number of locations to open is large,
        internal data structures in memory is large, but there is only a limited number
        of open files/task (limited by number of physical sion files)


