Each array is a folder that contains an array schema file (with information describing the array dimensions and attributes with their data types, along with other important metadata), and a set of timestamped subfolders called fragments. Each fragment corresponds to a batch write to the array, which can happen in any subdomain of the entire array domain (i.e., a fragment does not need to contain cells for the entire array domain). In addition, a fragment may correspond to a set of consolidated fragments, an important operation for optimizing performance in certain applications, described later. The fragment name consists of a random UUID, a timestamp range (in ms elapsed since epoch) and the format version of the fragment (for backwards compatibility). Moreover, each fragment is accompanied by an empty “ok” file, which is important for (eventual) consistency, to be discussed later. The figure below depicts an example of a 2D dense array with two attributes and 2 fragments, along with its file structure.