You should make at least five replication. 

any justification for why need to do 5 replications ? 

if 5 replications, do that mean i have to carry out 405 runs?

Kazim Sari is right, for a Full Factorial Design you will have to do 21 runs for five replications of center point. 1/2 Factorial Design and five replications its about 13 runs. Five replications for corner point for a Full Factorial Design you will need to do 80 runs, and for a 1/2 Factorial Design 40 runs.

I understand that your design is of 34 = 3*3*3*3 (4 factors each at 3 levels). In this case if you are doing a full factorial design than you’ll have 81 factor combinations (test conditions); you are going to need at least 2 replicates for each combination (total = 81*2 = 162 observations) to have enough degrees of freedom (dof) to complete the analysis of variance and to test the effects of the main 4 factors (each with 3 dof), the effects of the 6 two-way interactions (each with 4 dof), the effects of the 4 three-way interactions (each with 8 dof), the effect of the four-way interaction (with 16 dof), and to estimate the error with 81 dof. If you use only one replicate for the 81 combinations (one observation for each) than you wouldn’t have enough degrees of freedom to estimate the error term; the total degrees of freedom that you will have with one replicate is 80, which equals the sum of the degrees of freedom of all the effects of the main factors and the interactions, and hence leave no degrees of freedom to estimate the error. With two replicates you will have a total of 161 degrees of freedom, which will give you an extra 81 degrees of freedom to estimate the error term (161- 80 dof for the sum of the effects  = 81 dof for the error term).

You might be able to use only one replicate (81 observations) by assuming that certain high order interactions are negligible and use their sums of squares and degrees of freedom to estimate the error. For example, in your design, if you assume that the four-way interaction could be neglected than you can use its sum of squares and its 16 degrees of freedom to estimate the error and complete the analysis of variance.

is that 162 runs consider too large to run?

Yes, with the full factorial design with two replicates for each of the 81 factor combinations you'll have a total of 162 observations (runs).

If this is too much than you probably should consider simplifying the design to have less number of runs. The design suggested in Paul Sanchez ‘s answer above is a very good choice; you can only replicate the center point (0 0 0 0) and run all other corner points only once.

Many interesting answers, I agree with all of them. 

Another issue is the warmup period, did you think about it? You need a warmup period at the biggining of your simulation.

Concerning the number of replications, 5 is the minimum, in some simulation software you have an analyzer of results to optimize the simulation time, the warmup time and the number of replications.

are they simulation runs? are your settings deterministic? 

You don't need to replicate if you can use some of higher-order interactions as errors (i.e. the effects are considered negligible). If you are not sure whether higher order interactions are negligible or not, then go for replication at one combination of factors.  

i thin it is ok, but with no replication, MSE will be error. so, i recommend to you to use bissel or fang to analysis for DOE with no replication

4 levels and 3 factors can you tell me the DOE matrix from CCD and OA method .