Ed,
Okay, now we are in full agreement. I corrected myself and addressed the
larger performance difference in another message that you may not have seen
yet.
--Dave
On Thu, Jun 2, 2016 at 3:13 PM, Ed Hartnett <edwardjameshartnett@xxxxxxxxx>
wrote:
> Yes, I get that, but in this case that should only be about an order of
> magnitude difference in access speed, since he is only reading 8 values
> (hitting the disk 8 times instead of the 1 time it would take for a
> variable that is not unlimited.)
>
> But he is describing a much larger difference in performance...
>
>
> On Thu, Jun 2, 2016 at 5:01 PM, Dave Allured - NOAA Affiliate <
> dave.allured@xxxxxxxx> wrote:
>
>> On Thu, Jun 2, 2016 at 2:41 PM, Ed Hartnett <
>> edwardjameshartnett@xxxxxxxxx> wrote:
>>
>>> I don't think the fact that it is an unlimited dimension should make it
>>> any slower, either in classic
>>>
>>
>> <snip>
>>
>> Ed, I have experienced this slow down many times with Netcdf-3 classic.
>> As I said earlier, if the format is classic, then a coordinate variable on
>> the unlimited dimension is scattered throughout the file. This is
>> confirmed in the Users' Guide. At the low level, many disk blocks must be
>> read to get the complete coordinate array.
>>
>> --Dave
>>
>
