In this section we describe several key properties of efficient
large-scale parallel MLDM systems addressed by the GraphLab
abstraction [24] and how other parallel frameworks fail to address
these properties. A summary of these properties and parallel frameworks
can be found in Table 1.