No abstract

Long flow detection and load balancing are crucial techniques for data center running and management. However, both of them have been independently studied in previous studies. In this paper, we propose a complete solution called Sonum, which can complete long flow detection and scheduling at the same time. Sonum consists of a software-defined synergetic sampling approach and an optimal network utilization mechanism. Sonum detects long flows through consolidating and processing sampling information from multiple switches. Compared with the existing prime solution, the missed detection rate of Sonum is reduced by 2.3%–5.1%. After obtaining the long flow information, Sonum minimizes the potential packet loss rate as the optimization target and then translates load balancing into an optimization problem of arranging a minimum packet loss path for long flows. This paper also introduces a heuristic algorithm for solving this optimization problem. The experimental results show that Sonum outperforms ECMP and Hedera in terms of network throughput and flow completion time.

Background The incidence of cardiac implantable electronic device infection (CIEDI) is low and usually belongs to the typical imbalanced dataset. We sought to describe our experience on the management of the imbalanced CIEDI dataset. Methods Database from two centers of patients undergoing device implantation from 2001 to 2016 were reviewed retrospectively. Re-sampling technique was used to improve the classifier accuracy. Results CIEDI was identified in 28 out of 4959 procedures (0.56%); a high imbalance existed in the sizes of the patient profiles. In univariate analyses, replacement procedure and male were significantly associated with an increase in CIEDI: (53.6% vs. 23.4, 0.8% vs. 0.3%, P < 0.01). Multivariate logistic regression analysis showed that gender (odds ratio, OR = 3.503), age (OR = 1.032), replacement procedure (OR = 3.503), and use of antibiotics (OR = 0.250) remained as independent predictors of CIEDI (all P < 0.05) after adjustment for diabetes, post-operation fever, and device style, device company. There were 616 under-sampled cases and 123 over-sampled cases in the analyzed cohort after re-sampling. The re-sampling and bootstrap results were robust and largely like the analysis results prior re-sampling method, while use of antibiotics lost the predicting capacity for CIEDI after re-sampling technique (P > 0.05). Conclusion The application of re-sampling techniques can generate useful synthetic samples for the classification of imbalanced data and improve the accuracy of predicting efficacy of CIEDI. The peri-operative assessment should be intensified in male and aged patients as well as patients receiving replacement procedures for the risk of CIEDI.