Elastic-DF: Scaling Performance of DNN Inference in FPGA Clouds through Automatic Partitioning

Customized compute acceleration in the datacenter is key to the wider roll-out of applications based on deep neural network (DNN) inference. In this article, we investigate how to maximize the performance and scalability of field-programmable gate array (FPGA)-based pipeline dataflow DNN inference accelerators (DFAs) automatically on computing infrastructures consisting of multi-die, network-connected FPGAs. We present Elastic-DF, a novel resource partitioning tool and associated FPGA runtime infrastructure that integrates with the DNN compiler FINN. Elastic-DF allocates FPGA resources to DNN layers and layers to individual FPGA dies to maximize the total performance of the multi-FPGA system. In the resulting Elastic-DF mapping, the accelerator may be instantiated multiple times, and each instance may be segmented across multiple FPGAs transparently, whereby the segments communicate peer-to-peer through 100 Gbps Ethernet FPGA infrastructure, without host involvement. When applied to ResNet-50, Elastic-DF provides a 44% latency decrease on Alveo U280. For MobileNetV1 on Alveo U200 and U280, Elastic-DF enables a 78% throughput increase, eliminating the performance difference between these cards and the larger Alveo U250. Elastic-DF also increases operating frequency in all our experiments, on average by over 20%. Elastic-DF therefore increases performance portability between different sizes of FPGA and increases the critical throughput per cost metric of datacenter inference.

Inter-Cluster Competition and Resource Partitioning May Govern the Ecology of Frankia

Microbes live in a complex communal ecosystem. The structural complexity of microbial community reflects diversity, functionality as well as habitat type. Delineation of ecologically important microbial populations along with exploration of their roles in environmental adaptation or host-microbe interaction has a crucial role in modern microbiology. In this scenario, reverse ecology (the use of genomics to study ecology) plays a pivotal role. Some studies have reported the presence of other non-Frankia genus from the same root nodule from where the Frankia was isolated. Since co-existance of two different genus in one small niche should maintain a strict direct interaction, it will be interesting to utilize the concept of reverse ecology in this scenario.Here, we exploited an ‘R’ package, the RevEcoR, to resolve the issue of co-existing microbes which are proven to be a crucial tool for identifying the nature of their relationship (competition or complementation) persisting among them. Our target organism here is Frankia, a nitrogen-fixing actinobacterium popular for its genetic and host specificity nature. According to their plant host, Frankia has already been subdivided into four clusters CI, CII, CIII and CIV. Our results revealed a strong competing nature of CI Frankia. The competition index between CI and CIII was greater than other studied Frankia clusters. The other interesting result was the co-occurrence of C-II and C-IV groups. It was revealed that these two groups follow the theory of resource partitioning in their lifestyle. Metabolic analysis along with their differential transporter machinery validated our hypothesis of resource partitioning among C-II and C-IV group.

Interactive effects of light, CO2 and temperature on growth and resource partitioning by the mixotrophic dinoflagellate, Karlodinium veneficum

There is little information on the impacts of climate change on resource partitioning for mixotrophic phytoplankton. Here, we investigated the hypothesis that light interacts with temperature and CO2 to affect changes in growth and cellular carbon and nitrogen content of the mixotrophic dinoflagellate, Karlodinium veneficum, with increasing cellular carbon and nitrogen content under low light conditions and increased growth under high light conditions. Using a multifactorial design, the interactive effects of light, temperature and CO2 were investigated on K. veneficum at ambient temperature and CO2 levels (25°C, 375 ppm), high temperature (30°C, 375 ppm CO2), high CO2 (30°C, 750 ppm CO2), or a combination of both high temperature and CO2 (30°C, 750 ppm CO2) at low light intensities (LL: 70 μmol photons m-2 s-2) and light-saturated conditions (HL: 140 μmol photons m-2 s-2). Results revealed significant interactions between light and temperature for all parameters. Growth rates were not significantly different among LL treatments, but increased significantly with temperature or a combination of elevated temperature and CO2 under HL compared to ambient conditions. Particulate carbon and nitrogen content increased in response to temperature or a combination of elevated temperature and CO2 under LL conditions, but significantly decreased in HL cultures exposed to elevated temperature and/or CO2 compared to ambient conditions at HL. Significant increases in C:N ratios were observed only in the combined treatment under LL, suggesting a synergistic effect of temperature and CO2 on carbon assimilation, while increases in C:N under HL were driven only by an increase in CO2. Results indicate light-driven variations in growth and nutrient acquisition strategies for K. veneficum that may benefit this species under anticipated climate change conditions (elevated light, temperature and pCO2) while also affecting trophic transfer efficiency during blooms of this species.

Resource Partitioning of Scots Pine (Pinus sylvestris L.) by Pine Shoot Beetles in Stands under Stress Conditions

The pine shoot beetles Tomicus piniperda L. and T. minor Hartwig are sympatric species that occur on Scots pine in two habitats. Feeding by the beetles in tree crowns causes significant losses in tree growth and disturbs the crown’s proper development. A review of the subject literature showed that there had been no previous studies of interspecific competition in stands with different degrees of crown damage. The aim of this work was to assess the resource partitioning of stems by the two species in stands with damaged and undamaged crowns. Data were collected in the years 1992–2008 in stands containing Scots pine located at different distances from timber yards. A total of 259 natural traps were laid, and measurements of height and diameter at breast height were made for 900 pines. The surface area of each stem was divided into 20 equal sections by making a division lengthwise (into units) and laterally (into an upper and lower part). In total, 90,501 egg galleries of pine shoot beetles were counted on 9560 stem sections. Feeding by pine shoot beetles in the crowns of pines reduces site productivity and the nutritional suitability of stems. The results of niche segregation indicate pine shoot beetles exhibited spatial specialization in the use of resources. prefers the thicker part of the stem, and T. minor the thinner part. The population of T. piniperda on the trap logs was described using a multiple linear regression model with three explanatory variables. As a result of regression modelling, from the set of variables representing characteristics of habitats, trees and trap logs and the parameters of infestation, the following explanatory variables were selected: range of colonisation of a trap log (rc), site quality class (sqc), and crown undamaged (cu). The explanatory variables included in the MLRM model explain to a significant degree (p < 0.05) the niche breadth of T. piniperda on trap logs. In all validated plots, the mean real and model values for the niche of T. piniperda on the trap logs are similar (p > 0.5), confirming the high accuracy of the developed model.