This is the manual for older MicroStream versions (Version < 5.0).
The new documentation (Version >= 5.0) is located at:
Yes. In fact, every storing of data is executed as a transaction, an atomic all-or-nothing action. When one or more entities are stored, their data is collected into a continuous block of bytes and that block is written to the physical form (the "files") in one fell swoop. Any problem during the IO-operation causes the whole block to be deleted (rolled back).
Yes. The storing and loading process can be parallelized by using multiple threads and thus be strongly accelerated. There is no limitation on how many threads can be used, apart from the mathematical constraint that the thread count must be a power of 2 (1, 2, 4, 8, 16, etc.).
Yes. There are currently two options available to create backups: An on-the-fly backup that copies and validates stored entity data after it has been written and the possibility to export all database files to a target location (which is in fact just a low-level file copying, but executed in a concurrency-safe way).
Yes. MicroStream provides a per-type export of binary data and a utility to convert its binary data into the CSV format. The other way around (convert CSV to binary an import binary files) is also possible.
No, because it doesn't need to. Such concerns are long covered by the application itself, with the DBMS usually being degraded to only being the application's exclusive database. Thus, all that is needed for a modern business application is just an application-exclusive data storage solution, which is exactely what MicroStream is.
Yes, if the data is structured in a format conforming to the entity classes and with references being represented in globally unique and bijective numbers. How hard that is for a given database depends on its specifics, but it can be as easy as executing one generically created SELECT per table.