[GRASS-dev] graph function input limitations

[Paulo van Breugel]

On Wed 21 Nov 2012 03:39:44 PM CET, Glynn Clements wrote:
>
> Paulo van Breugel wrote:
>
>> I am working with the r.mapcalc graph function. When the input (number
>> of xy pairs) is very high, I am getting an error message:
>>
>> "memory exhausted
>> Parse error
>> ERROR: parse error"
>>
>> The largest possible number of xy pairs seems to be somewhere between
>> 2400 and 2500. It runs with 2400 xy pairs, and pretty fast for that
>> matter. Memory usage doesn't go beyond 1.5GB with 2400 xy pairs, while I
>> have 12GB RAM (on 64bit Linux computer), so I guess RAM isn't the
>> limitation here.
>
> Each "term" in an r.mapcalc expression requires a row buffer, i.e. one
> int/float/double value for each column of the map. A numeric literal
> results in a row buffer filled with that value.
>
> Depending upon what you're trying to do, it might be possible to use a
> combination of r.mapcalc -> r.recode -> r.mapcalc. r.recode would use
> significantly less memory than r.mapcalc's graph() function.

Thanks for the suggestion, I will see if I can find a way to use 
r.recode

>
>> In the help page of r.series, there is a mention of a soft limit of 1024
>> files. Does something similar apply here (but with different numbers)?
>>
>> I can actually work around it, using a approximation of the graph
>> function, but would like to know at what number of xy pairs I should
>> apply the workaround. Particularly important, is the limit the same for
>> all users, or if not, what would be a safe limit? Or is there a way to
>> set this limit from within a script?
>
> The limit would depend upon the number of columns in the current
> region and the amount of memory available.

OK, but what I find strange is that using 2400 terms uses less then 2.5 
GB, I would not expect 2500 terms to use more then 12 GB (amount of RAM 
available) in that case.
I also tried to run a r.mapcalc with 2600 terms for a map of 15 x 19 
cells. Also here I got the message: "memory exhausted", which can 
hardly be related to lack of memory?


>