> I read a single (lat, lon) point from a file at a time for the whole
available time period
Yes, that pattern is why your reads are so slow. Look at how the data is
laid out:
float u10m(time=5088, latitude=103, longitude=122);
So, if you're reading all values for a single (lat, lon) point, you are
accessing the array like so:
u10m[0][lat][lon]
u10m[1][lat][lon]
u10m[2][lat][lon]
...
u10m[time-1][lat][lon]
Those values are physically located far apart within the file, preventing
effective buffering due to the non-sequential reads.
Chunking won't help you here; you'll need to select the data layout that
suits your read pattern. This might work better:
float u10m(latitude=103, longitude=122, time=5088);
Cheers,
Christian
On Wed, Feb 18, 2015 at 11:51 PM, Antonio Rodriges <antonio.rrz@xxxxxxxxx>
wrote:
> P.S.
>
> And that speed (~700 ms) is both for chuncked and unchunked data so I
> decided that is was "slow" for the chunked file since it is considered
> more optimized for this read pattern.
>
