HLA-based simulation system, regarded as a special kind of distributed system, often adopts rollback recovery to realize fault tolerance. Checkpoint interval is an important rollback recovery parameter that will seriously influence system performance. Firstly, we analyze the differences of fault tolerance between HLA-based distributed simulation system and the general distributed system. And then according to the different degrees of the importance of the simulation process to the simulation result, we classify simulation processes into trivial parts and critical parts. Furthermore, the availability of distributed simulation system, which adopts rollback recovery mechanism, has been defined and analyzed through the utilization of Markov chain. As a result, we achieve an equation, by which the checkpoint interval in the best system availability can be figured out. The correctness of this conclusion has also been testified through a set of experimental data.