Big Data Computing

283Big Data Exploration

When cracking physically reorganizes one column, the rest of the columns

of the same table remain intact; they are separate physical arrays. As a result,

with cracking, columns of the same table are not aligned anymore. Thus,

when a future query needs to touch more than one columns of the same

table, then the system is forced to perform random access in order to recon-

struct tuples on-the-y. For example, assume a selection on a column A, fol-

lowed by a projection on another column B of the same table. If column A

has been cracked in the past, then the tuple IDs, which is the intermediate

result out of the select operator on A, are in a random order and lead to an

expensive access to fetch the qualifying values from ...