Hash memory usage with replace() method in SAS - hash

Question

Why does my hash exceed the memory limits when I use the replace() method, when if I use the same code without the replace method the hash fits just fine? It seems like the hash would remain the same size either way. I am running the code on unix. In the code below, if I comment out ht.replace() the code runs fine. If I leave it in (don't have it commented out) then I receive a message saying "Hash object added 2490352 items when memory failure occurred." The "series" data set which is fed into the hash has 13 variables and 6912 rows. The "data1" dataset has 26970 rows and 4 columns. Is there any way to resolve this without messing the memsize?
data _null_;
if 0 then set series;
if _n_ = 1 then do;
declare hash ht(dataset:"series", ordered:"a", multidata:"yes");
rc = ht.defineKey("one", "two", "three");
rc = ht.defineData(all:"yes");
declare hiter hi("ht");
rc = ht.defineDone();
end;
set data1 end=eof;
rc = hi.first();
do while (rc = 0);
if low <= code1 <= high then do;
sum = sum + value1;
ht.replace();
end;
rc = hi.next();
end;
if eof then ht.output(dataset:"sum1");
run;

Answer 1

Probably, the problem is that your hash is multidata one, i.e. one key can correspond to many data items. For multidata hashes you have to use REPLACEDUP-method, unambiguously selecting not only a specific key, but also a specific data item within this key.
So your iterating over hash ht should look like this:
rc = hi.first();
do while (rc = 0);
rc=ht.find_next();
do while(rc=0);
if low <= code1 <= high then do;
sum = sum + value1;
ht.replacedup();
end;
rc=ht.find_next();
end;
rc = hi.next();
end;