[Orca-dev] Orca 0.26 released

[Blair Zajac]

Hello,

I've just finished the 0.26 version of Orca.  This has a lot
of work put in it, including notable performance enhancements
and some major code restructuring to make it easier to
understand and edit.

You can get it from

http://www.gps.caltech.edu/~blair/orca/

Here's the change list from the previous release.

Blair


 1) Due to some changes at onelist.com, the name of this mailing must
    change.  The -help at the end of orca-help is being reserved for
    the mailing list named "orca".  As such, I'm renaming this list to
    orca-users at onelist.com.

 2) Major speedup in starting Orca when many different source files
    exist and the first line of the file is used to tell Orca what
    columns of data there are.  Now Orca caches this information in
    the state file so that the source data files do no have to be
    opened everytime Orca starts up.  This has a significant speed
    improvement when the file is compressed, since a separate process
    is normally spawned to read the file.

 3) Reduce the number of size of the open file descriptor cache to 100
    open file descriptors since the change in 2) reduces the need for
    as many open file descriptors.

 4) A new configuration file parameter named filename_compare was
    added that let's Orca determine when it can flush data from loaded
    source files.  This prevents Orca from loading in all of the data
    and then flushing it to the disk when the amount of new data is
    large.  See the Orca documentation for more information.

 5) Allow Orca to work with RRDtool versions newer than 1.0.7.

 6) Save Orca's state more often so that if something happens to the
    Orca process, it does not have to reload all the data from the
    source data files it previously read.

 7) Have Orca print the current memory usage and CPU time more often.
    This attempts to use the proper ps command for the operating
    system Orca was configured on.

 8) Take all of the modules in the monolithic orca script and move
    them into their own files.  This improves readability and
    maintainability.

 9) Require the latest versions of Data::Dumper (2.101), Digest::MD5
    (2.09), Math::IntervalSearch (1.05), RRDs (1.0.13), and Storable
    (0.6.9).

10) Use arrays instead of hashes as the internal structure for many
    objects.  This should speed access to object members and make
    their memory footprint smaller.

11) Use a new integer file identifier (FID) instead of the filename to
    refer to in internal objects.  There is now a global array keyed
    by the FID that gives the name and a hash keyed by filename that
    returns the FID.  The FID is used where the filename previously
    was used.  This reduces Orca's memory usage by 10%, since only one
    copy of a long filename is needed.

    These following changes are what's new in orcallator.se 1.23 and
    the orcallator.cfg file since version 1.20 which was included with
    Orca 0.25.  All of the changes below are taken advantage of in the
    included orcallator.cfg and start_orcallator files.

12) Include code to record and plot the NFS v2 and v3 server
    statistics.  The new statistics are: nfss_calls, the number of NFS
    calls to the NFS server, nfss_bad, the number of bad NFS calls per
    second, and v{2,3}{reads,writes}, which are nfss_calls broken down
    into NFS version 2 and NFS version 3 calls.  The sum of
    v{2,3}{reads,writes} will be less than nfss_calls as the other
    types of NFS calls, such as getattr and lookup, are not included.
    Contributed by Paul Haldane <Paul.Haldane at newcastle.ac.uk>.  This
    code is enabled by the standard -DWATCH_OS or individually by
    -DWATCH_NFS_SERVER.  The define -DWATCH_NFS has been supperseded
    by -DWATCH_NFS_CLIENT, but to keep backwards compatibility,
    -DWATCH_NFS_CLIENT will be defined if -DWATCH_NFS is defined.

13) When orcallator was running on a system with DiskSuite, the same
    physical disk was listed multiple times when it appeared in the
    same metadevice.  The solution to the problem is not to build the
    c0t0d0 name but use the long disk name provided by the long_name
    string. Patch contributed by Paul Haldane
    <Paul.Haldane at newcastle.ac.uk>.

14) Prevent core dumps on extremely long access log lines.