Deep learning-based recommender models (DLRMs) have become an essential component of many modern recommender systems. Several companies are now building large compute clusters reserved only for DLRM training, driving new interest in cost- & time- saving optimizations. The systems challenges faced in this setting are unique; while typical deep learning (DL) training jobs are dominated by model execution times, the most important factor in DLRM training performance is often online data ingestion.

A mobile ad-hoc network (MANET) is a collection of wireless autonomous nodes without any fixed backbone infrastructure. All the nodes in MANET are mobile and power restricted and thus, disconnection and network partitioning occur frequently. In addition, many MANET database transactions have time constraints. In this paper, a data replication technique for real-time ad-hoc mobile databases (DREAM) is proposed that addresses all those issues. It improves data accessibility while considering the issue of energy limitation by replicating hot data items at servers that have higher remaining power. It addresses disconnection and network partitioning by introducing new data and transaction types and by considering the stability of wireless link. It handles the real-time transaction issue by replicating data items that are accessed frequently by firm transactions before those accessed frequently by soft transactions. DREAM is prototyped on laptops and PDAs and compared with two existing replication techniques using a military database application. The results show that DREAM performs the best in terms of percentage of successfully executed transactions, servers' and clients' energy consumption, and balance of energy consumption distribution among servers

A Mobile Ad-Hoc Network (MANET) is a collection of wireless autonomous mobile nodes with no fixed infrastructure. Since no fixed infrastructure is required, MANET fits well in military operations, emergency disaster rescue, and mobile ad-hoc voting. There are many issues that have to be addressed while designing a technique for managing real-time database transactions in MANET: 1) energy limitations; 2) client and server mobility; 3) real-time constraints imposed on transactions; and 4) frequent disconnection and network partitioning. We have designed PETRANET1: a Power-Efficient Transaction management technique for Real-time mobile Ad-hoc NETwork databases that addresses the above specified issues. In this paper, we present a system prototype that we have developed to implement PETRANET for a military database application.

In a Mobile Ad-hoc Network (MANET), due to the mobility and energy limitations of nodes, disconnection and network partitioning occur frequently. In addition, transactions in many MANET database applications have time constraints. In this paper, a Data REplication technique for real-time Ad-hoc Mobile databases (DREAM) that addresses all these issues is proposed. DREAM is prototyped on laptops and PDAs and compared with two existing replication techniques using a military database application.