Prevent simulation from being killed by OS

Question

This is not homework, just a learning exercise on my part. I have been running a very simple simulation (or number-crank) in R. It generates two numbers (A, B) and runs for 1 month.

A=NULL
B=NULL
x=NULL
x <- Sys.time()
duration <-  2592000 # 30 days
while(Sys.time() <= x + duration){
A <-append(A, sample(1:5, 1000, 1/5))
B <-append(B, sample(1:5, 1000, 1/5))
save.image()
}

I thought it was going well, but after one week (and several million numbers generated) the OS killed the process. Is there a better way of writing or running the simulation that would prevent the OS killing it?

I would prefer to rewrite the simulation than to adapt the OS (such as adding more swap etc). I am running the simulation on a low-powered device (Raspberry Pi) and am limited in what I can do on the hardware side. Thanks.

UPDATE:
1) It is not important that the samples be generated 1000 at a time. This was just my cludge.
2) It is important that the simulation runs for a set period of time ie 1 week, 1 month or 1 year.
3) Unless impossible I want the raw data.

Answer 1

If the goal is to create two big samples, consider the following:

N <- 2000000
A <- sample(1:5, N, 1/5)
B <- sample(1:5, N, 1/5)
save.image()

If it is important that the samples be formed 1000 at a time for A and B alternatively, consider this:

N <- 2000
n <- 1000
A.list <- vector("list", N)
B.list <- vector("list", N)
for (i in 1:N) {
   A.list[[i]] <- sample(1:5, n, 1/5) 
   B.list[[i]] <- sample(1:5, n, 1/5)
}
A <- unlist(A.list)
B <- unlist(B.list)
save.image()

This should take care of the two main issues in your code:

• every time you use append inside your loop, R has to create and fill a couple new objects from scratch. As the objects become larger, your loop iterations becomes slower and slower; computation times grow quadratically I believe. You also run the risk of fragmenting your memory space, this is harder to explain but you can try to research it. By using a list, only the new data from each iteration needs to be stored to memory and the computation time per loop remains the same.
• I have moved save.image() outside of the loop. Same idea, saving objects as they get bigger and bigger will take longer and longer, i.e. slow down your iterations. Since you only care for the final vector, it makes sense to only save when you are done.

You can play with the value of N to see how far your OS will let you go. The advantage is that you don't have to wait for a week or a month to find out what the limits are.

Answer 2

If you consider printing out the result on paper as an acceptable solution, then Roman Luštrik's solution (in the comments to your question) of appending your data to a text-file or a database is definitely one good solution.

Here is what appending to a text-file would look like:

x <- Sys.time()
duration <-  2592000
while(Sys.time() <= x + duration){
    write.table(sample(1:5, 1000, 1/5),file="A.txt",append=TRUE,row.names=FALSE,col.names=FALSE,sep="\t")
    write.table(sample(1:5, 1000, 1/5),file="B.txt",append=TRUE,row.names=FALSE,col.names=FALSE,sep="\t")
}