A model that can fit binary data in R I have data that indicates if the computers in my network are up or down. Data is collected every minute and has a seasonality of one week, looks like this for 3 hosts.
Servers are supposed to go down less often than workstations.   
ID TIME STATE TYPE  
1  60   0     workstation  
2  60   1     server  
3  60   0     server  
1  120  1     workstation  
2  120  1     server  
3  120  0     server  

I am using R and I'm looking for a model that can give me the probability of a host being down or up in a future time interval.
Which model do you think would fit the best these requirements?
I have thought of using duration models with the survival package, considering the host going up and the host going down as events.
What do you think of this ? Are there better candidates?
 A: You are right, survival models would be one choice. These estimate the probability to remain "up" for another period (minute), given the covariates. I guess the only covariate here would be a constant and a workstation dummy. This makes the most sense if computers remain "down" once they are down absent intervention.
Another way that makes more sense if computers go up or down randomly: OLS with autoregressive terms. If a station is down at $t$, it will likely also be down at $t+1$, right? Hence, something like 
$$state_{i,t}=\alpha+\beta_1 state_{i,t-1}+\beta_2 state_{i,t-2}+\beta_3 state_{i,t-3}+\beta_4 workstation+e_i$$ 
might be appropriate (workstation is dummy). Perhaps also add interactions of autoregressive terms with workstation. This would allow you to predict the probability of being up or down in the next minute. 
If you want to predict the probability of the state in two minutes, you have to respecify and reestimate the model dropping the $state_{i,t-1}$ terms (since you don't know what the state in the period prior---that is the next---period will be). And so on.
