Dear Authors,

although you answered the questions by the referees in detail,
the manuscript is not revised sufficiently, because important information
from the discussion should be included in the final manuscript
(which must stand as its own). As such, please

1) cite the manuscript of Palmera et al., 2011 (and maybe others)
"stressing the importance of utilizing data aggregation to achieve
high bandwidth" as motivation of your work at a suited position
in your text (maybe introduction).

2) add references for the netcdf and pnetcdf libraries (at least the
URLs, or even better papers, if available).

3) explicitely mention (e.g., in Section 3) which type of pnetcdf interface
you used: collective parallel netcdf API.
(see your reply on page C3216)

4) add a discussion section, where you elaborate on important topics:
a) The issue on I/O performance tuning (with reference to
Behzad and Lu, 2013) for application scientits and the
limitation that you did not use such tuning (stripe count, stripe size)
(see comment/reply on page C3219/C3220).
This limitation needs to be explicitely mentioned.
b) Elaborate a bit on the pros and cons and the limitations of
application level versus software stack level data aggregation
techniques (comments/replies on pages C3215, C3220/C3221)

5) state, e.g. in Section 4.2 that the simplified code is available on
request

Moreover, I think there was a misunderstanding of one referee comment (see
page C3215, last paragraph and subsequent page). As far as I understand the
referee an argument on your text is oversimplified. The referee states:

"In section 5.3 the authors claim that increasing the data size on the
processor which is responsible for I/O will translate into higher I/O
rates. If this is the case why not just use the original serial approach in
which one I/O processor is responsible for all of the data?"

So, please elaborate on the question, why this simple connection
(increased data size -> higher I/O rate) does obviously not hold
completely (otherwise serial I/O should be the fastest).
Please modify your text/argumentation accordingly.

To facilitate the further review and editorial process, I would
very much appreciate, if you could provide me a revised document
in which the modifications are highlighted, but only if this is
technically possible for you. In case not, please provide
a detailed list of modifications with (page/line numbers).

Sincerely,

Patrick Jöckel