The following R  code is sufficient to replicate the results in table 1:
```{r}

#minimum reported failure rate in memory is .2 per errors GB/day

rate_GBday_wikipedia <- 0.2

# assumed quantity that new data remains resident in memory for 1 second

newdata_Gb_sec <-  1E9

# p value for randomized response from Wang et al. 2016
randp <- function(eps) {
  exp(eps)/(1+exp(eps))
}

# predicted bit errors per day
dayerr <- function(eps,
                   flow = newdata_Gb_sec  ) {
  res <- (1-randp(eps))  * flow * 60 * 60 * 24
  res
}

# find epsilon with  5% change of observed error in an hour
optim(par=8, fn=function(x){abs(.05-dayerr(eps=x)/24)}, method="Brent", lower=.1, upper=42)

# find epsilon for a given bits/day error rate
optim(par=8, fn = function(x) { abs(rate_GBday_wikipedia - dayerr(eps=x))}, method="Brent", lower=.1, upper=42)
``` 


The following code was used to to conduct the experiment reported in section 7:
```{r}

blockSize <- 200000000
simmem <- integer(length=blockSize)
simmem[1:length(simmem)]<-as.integer(0)
if (object.size(simmem)/1000000 < 50)
   {warn("test vector may be smaller than processor cache")}

startTime <- Sys.time()
 simsum <- 0
while (simsum==0) {
 # alter calculation to avoid optimizer converting entire loop to while(TRUE)
  oneout <- trunc(runif(1,1,length(simmem)))
  simsum <- sum(simmem[-oneout])
}
endTime <- Sys.time()
elapsed <- endTime-startTime
print(paste("Time",endTime ,"sum", simsum, "elapsed", elapsed))

```