Fast and easy access to structured data stored in memory-mapped files

Most well-known structured data serialization libraries possess a dedicated

deserialization routine, which parses persistent data on disk into

its in-memory form. This routine might be both CPU- and memory-expensive,

and while most of the time it isn't a problem, there is a couple

of cases where it is desirable to eliminate this step completely --

either because of constraints on application startup time or just because

deserialized data might not fit into RAM.


For such cases we've adopted another approach, storing our structured data

in such form that we can just mmap() them and then use them, without

any additional deserializaton step. Besides, our approach does not impose

any constraints on data complexity and provides a nice STL-like interface

to containers.


In the talk we'll present key concepts of this method, examine user data

and standard containers serialization routines in more detail and talk

about method's pros and cons.

Speaker: Dmitry Prokoptsev

Go back