Re: Veritas NetBackup.

[prev] [thread] [next] [lurker] [Date index for 2006/11/28]

From: Darrell Fuhriman
Subject: Re: Veritas NetBackup.
Date: 03:56 on 28 Nov 2006
> Man, I'd forgotten all that robot-inventorying nonsense.  Years of
> therapy, all gone down the tubes now.  Alas.

Alright, I feel I must whip out my 24 point "Why Netbackup sucks"  
rant that I sent to our SE in... wait for it... 2000. Sounds like  
little has improved.

Thanks for reviving bad memories.

Darrell

----------

All right, Dave, you guys asked for it.

I've drawn up a list of things that bug me, in no particular order.
Undoubtedly there are more, but this a good start.  :)  Things marked
with * are things that make me say, "What a FPOS.  Can we use something
else that's not so stupid?").

This list only took me about 30 minutes to draw up off the top of my  
head,
that should tell you something.

I'll be happy to discuss any of these in more detail if you have
questions.

Darrell

1) netbackup needn't send a notice for every failed job.
    I'd rather see a summary of failed jobs for each host when the  
window
    closes or when all jobs are sucessful. Information on whether
    a job sucessfully retried should be included.

* 2) ability to create multiple default exclude/include lists on a  
master
    and assign them on per client/class basis.  i.e.  not have to  
push them
    out from the master.  local include/exclude lists should be  
additive,
    not substitutive to the master list.

3) the ability to define administrators.  I don't want everyone
    who wants to change the system have to have root access. Ideally,
    there would also be a 'read only' view, and an operator
    view.  Operators should be able to change tapes and the like, but  
not
    mess with the configuration.

* 4)
# ./bpgp
Segmentation fault

No comment necessary.

* 5) every single command should have a complete man page.
    I don't want to hear crap like "Users shouldn't
    be using that command."  Every single command line function should
    return accurate usage information with a -help (or equiv) flag.
    Every single config file should have also have a complete  
manpage.  By
    this, I don't mean merely bp.conf and the like, but job queue files,
    etc. etc.

    This near complete lack of documentation is inexcusable.

6) calendar based scheduling

* 7) flat text indexes?  come on, what is this? 1980?

* 8) when a tape expires, the indexes should not be deleted.  The
    tape should merely be marked as available for reuse.  No information
    should be deleted until the tape is actually reused.

9) nbar window usage graph should show when a window closed, even
    if the window is exceeded.  That is non-obvious in the current  
version.

10) a report on library usage, including duplicates.  I want to be able
     to see how close I am to filliing a given library

11) nbar should have ability to have thresholds configured for some
    reports and generate a report on exceeded thresholds.  (For  
instance,
    I'd like to be able to get a report that tells me if the library in
    (10)  has exceeded 90% usage, or I'd like to know everytime a backup
     window is exceeded.)  Think a less real-time version of SNMP traps.

12) support for ndmp 3/4 tape positioning information.

* 13) I would absolutely love to use TIR, make it not suck.

14) ndmp should be able to restores as flexibly as regular backups.

* 15) all newly added hosts should get a full backup as part of their
     first run, regardless of what the scheduler says.

16) an equivalent to TSM all-incremental backups would be really, really
     nice. that may require to-disk staging, but that's fine with me.

17) using the TIR information, I would like to be able to make a
     'snapshot' of a particular point in time, and copy that snapshot to
      another tape.
     i.e. make an arbitrary archive out of the backup information  
already
     stored

18) use a tape format that allows positioning to individual files on
     the tape.  (disk files, not tape files).  I wanted finer grained
     access than merely fragment size.  This would speed restores
     significantly.

19) large job queues make the Job Monitor run slow, slow, slow.

20) proper package based install, and patching.  Every single command
     should be able to return a version string, including patch level
     in a standard fashion.  (-version)

* 21) the ability use ssh instead of rsh for installs.  But really,
    the client software shouldn't have to be pushed from the server
    at all.  All supported client platforms should have a package  
available
    for install that is appropriate to the platform.  i.e. RPMs for
    Linux, pkgadd for Solaris, etc.  rsh is dead.  get over it.

22) I should be able to assign as class to multiple tape libraries, not
     just one or 'any available'.

23) networker .nsr files are *really* handy.  make something equivalent.

* 24) the networker 'recover' command is really, really excellent.
       steal it.  I'm dead serious -- I want that command.

There's stuff above here

Generated at 16:01 on 28 Nov 2006 by mariachi 0.52