Define "arrival rate".
Transactions per second can be calculated like this:
data intps;
input hh mm ss;
datalines;
3 59 59.807316959
3 59 59.813843966
3 59 59.823140979
4 0 0.123
4 0 0.345
;
data transactions;
set intps end = done;
retain hour min sec count;
sec = floor(ss);
if hour ne hh or min ne mm or sec ne floor(ss)
then do;
if count > 0 then output;
count = 0;
hour = hh;
min = mm;
sec = floor(ss);
end;
count + 1;
if done then output;
keep hour min sec count;
run;
proc summary data=transactions;
by hour;
var count;
output out=want (drop=_type_ _freq_) mean()=avgtps max()=peaktps;
run;
Note that I use the rather complicated data step to avoid using SQL or an extra step that calculates the second without the fractions. You could replace it by
proc sql;
create table transactions as
select
hh,
mm,
floor(ss) as ss,
count(*) as count
from intps
group by hh, mm, calculated ss
;
quit;
and then use hh as BY variable in the SUMMARY. Which method you use should be determined by test runs with your production data.
You may also contemplate some method to insert "0" observations for missing seconds, so that you always have 3600 observations per hour.
