broken pipes (was Re: Performance)

 Sandro writes:
> Broken pipes were appearing in many places as I'd mentioned(ie. when
> compiling, untarring and gunzipping, etc.), so I thought it might be a
> problem. But if it's harmless, as some of you have said, then I'll leave it
> at that. I'm still wondering where they're all coming from though...

Pardon this intrusion from the peanut gallery (I don't actually run vsta, I
just lurk) but I have some general points that I haven't seen raised yet:

A ``broken pipe'' error comes when the writer writes to a pipe that has
already been closed by the other end. In this case the writer is most
likely the gzip spawned by tar, and the reader is tar. So most likely tar
is seeing an eof marker within the data before the data stream is actually
exhausted and is closing its end before gzip has finished fully writing the
last block.

The explanation could be as simple as tar not being compiled with the same
default block size on the system where the file was created. Or if not
that, it may be that vsta's scheduler is more agressive: rather than
letting gzip run for a while as a unix system would it passes control to
tar as soon as partial pipe write is complete (just another guess).

I suggest playing with tar options: try -B (reblock input), -i (ignore
zeros as eof marker) and/or -b N (to change the expected block size). If
these work then you don't have an OS problem, but maybe tar should be
compiled differently.

Cheers,
Jaime
...........................................................................
: James da Silva <jds@saltywaters.org> : Stand on my shoulders, :
: at work <jds@photuris.com> : not on my toes. :
Received on Wed Dec 27 08:24:41 2000