As with most SAS modeling procedures, the memory and time needed by CATMOD increase as the number of model parameters increases. If the number of parameters exceeds 100, CATMOD will print a NOTE on the SAS log telling you the number of parameters in the model and warning that the time and memory requirements could be very large. Generally, the number of model parameters goes up as the number of distinct values in the response variable and in each of the predictor variables goes up. Because of the types of models that CATMOD fits, the response variable should always have very few distinct values. If any of your predictors is continuous rather than discrete, you typically should specify it on the DIRECT statement. While CATMOD treats variables in the DIRECT statement in a continuous fashion, the procedure was designed for categorical predictors rather than truly continuous predictors which tend to have a unique value in each observation.
Some things to consider in reducing the time and memory requirements are either using a different procedure or reducing the number of parameters in your model. If you are using CATMOD to fit a binary or nominal multinomial logistic model, you should instead use PROC LOGISTIC or PROC GENMOD. If you are using CATMOD to fit a loglinear model, consider doing this in PROC GENMOD by modeling the cell counts with a log-linked, Poisson model. If you are primarily interested in assessing the association of one predictor with the response while controlling for some additional variable(s), then consider the CMH option in PROC FREQ, which doesn't require fitting a model.
Otherwise you may be able to reduce the number of parameters in CATMOD by: