In this paper we evaluate the effectiveness of Random Early Detection (RED) over traffic types categorized as nonadaptive, fragile and robust, according to their responses to congestion. We point out that RED allows unfair bandwidth sharing when a mixture of the three traffic types shares a link This urlfairness is caused by the fact that at any given time RED imposes the same loss rate on all jlows, regardless of their bandwidths.We propose Fair Random Early Drop (FRED), a modified version of RED. FRED uses per-active-jlow accounting to impose 011 each flow a loss rate that depends on the flow's buffer use.We sl~ow that FRED provides better protection than RED for adaptive uragile and robust) flows. In addition, FRED is able to isolate non-adaptive greedy trafic more effectively. Finally we present a "two-packet-buffer" gateway mechanism to support a large number of f7ows without incurring additional queueing delays inside the network These improvements are demonstrated by simulations of TCP and UDP traffic.FRED does not make any assumptions about queueing architecture: it will work with a FIFO gateway. FRED's peractive-jlow accounting uses memory in proportion to the total number of b@fers used: a FRED gateway maintains state only for flows for which it has packets buffered, not for all flows that traverse the gateway,

Learning e ective feature crosses is the key behind building recommender systems. However, the sparse and large feature space requires exhaustive search to identify e ective crosses. Deep & Cross Network (DCN) was proposed to automatically and e ciently learn bounded-degree predictive feature interactions. Unfortunately, in models that serve web-scale tra c with billions of training examples, DCN showed limited expressiveness in its cross network at learning more predictive feature interactions. Despite signi cant research progress made, many deep learning models in production still rely on traditional feed-forward neural networks to learn feature crosses ine ciently.In light of the pros/cons of DCN and existing feature interaction learning approaches, we propose an improved framework DCN-M to make DCN more practical in large-scale industrial se ings. In a comprehensive experimental study with extensive hyper-parameter search and model tuning, we observed that DCN-M approaches outperform all the state-of-the-art algorithms on popular benchmark datasets. e improved DCN-M is more expressive yet remains cost e cient at feature interaction learning, especially when coupled with a mixture of low-rank architecture. DCN-M is simple, can be easily adopted as building blocks, and has delivered signi cant o ine accuracy and online business metrics gains across many web-scale learning to rank systems.

No abstract

No abstract

No abstract