Gluster has a big users community and seems to be a good alternative to easily test any concept through its stackable translators approach. On the other hand, the current Distributed File Systems design has not really evolved and modern solutions still rely on an old metadata and I/O servers model. From our point of view, this model does not take into account the infrastructure’s physical topology, leading to expensive network communications that limit the performance and the scalability of the system. In this talk we'll discuss our impressions/advantages/problems of using Gluster when trying to develop a prototype that aims at reducing the traffic to the minimal needs and keep the applications performance over multiple sites.
