Dear Boost users/devs, Suppose i have a simulator S: R^n ---> R such that each run is during about 1 hour. Even using Boost.MPI, is expensive to run thousands of simulations all the time. To solve this problem, i'm wondering in use Boost.Serialization to store the pairs inputs/output in a boost::serialization::binary_archive and check before run if those inputs had an associated output already. By reading the Boost.Serialization docs, i can't found a way to deal with offsets in the archive, in really i don't know if it's possible. My intention is to get an archive like: 1.3, 2.5, ..., 4.6 0.1, 4.3, ..., 5.0 ... ... 123445 1234542 9893 1234 ... and skip the first rows, which are the inputs, to get the associated outputs. If you have any advice to solve this or know another more appropriate approach for this restart feature, please let me know. The second option i have in mind is work with C++ file streams and save the floating points by using the low-level std::fstream::read std::fstream::write functions. Best regards, Júlio.