New Jersey Institute of Technology

Huawei Technologies Co., Limited

Future Wei Technologies, Inc.

Yan Zhang

Nirwan Ansari

Hong Heather Yu

Yi Shan Jiang

Mingquan Wu

Chao Zhang

Junmin Zhao

Yongjian Chen

Nirwan Ansari Ansari

A compilation method includes obtaining a source program code. The source program code includes a first function in a first language code and a second function in a second language code. The first language code is a native language. The second language code is a non-native language. The method also includes generating a third language code based on the source program code. The third language code includes a third function, a fourth function and a fifth function. The third function is generated based on the first function. The fourth function is generated based on the second function. The fifth function is generated based on the first function and the second function. Executing the third function invokes the fourth function via the fifth function.

Cross-language compilation method and device

Technologies are generally described to allocation of virtual machines to physical machines through dominant resource assisted heuristics. According to some examples, multiple virtual machines (VMs) may be clustered to two or more unallocated VM clusters according to a dominant resource requirement associated with each of the VMs. The VMs may be sorted according to a size attribute associated with the dominant resource requirement. Multiple physical machines (PMs) may be sorted according to a power efficiency attribute associated with each of the PMs. One of the PMs may be selected from an ordered list of PMs based on the power efficiency attribute. One of the VMs may be selected from another ordered list of PMs based on the size attribute. The selected VM may then be allocated to the selected PM.

Allocation of virtual machines to physical machines through dominant resource assisted heuristics

Technologies are generally described to provide a congestion management system for a datacenter network. According to some examples, the congestion management system may detach management of congestion at the datacenter network from switches of the datacenter network. The congestion management system may also incorporate a distributed detection mechanism that is configured to detect a potential congestion culprit. Furthermore, detachment of the management of the congestion may allow the datacenter network to incorporate switches with limited traffic management capabilities. The switches may monitor a local congestion status and forward the congestion status to a centralized congestion controller that manages the congestion. The congestion management system may also be deployed for a cross-layer congestion management scheme.

Congestion management for datacenter network

Technologies are generally described for an enhanced Quantized Congestion Notification (QCN) congestion control approach, referred to as Fair QCN (FQCN) for enhancing fairness of multiple flows sharing link capacity in a high bandwidth, low latency data center network. QCN messages may be fed back to flow sources (e.g., servers) which send packets with a sending rate over their share of the bottleneck link capacity. By enabling the flow sources to regulate their data traffic based on the QCN messages from a congestion control component, the queue length at the bottleneck link may converge to an equilibrium queue length rapidly and TCP throughput performance may be enhanced substantially in a TCP incast circumstance.

Fair quantized congestion notification (FQCN) to mitigate transport control protocol (TCP) throughput collapse in data center networks

An embodiment of a system and method that uses inline measurements to probe available bandwidth for a transmission control protocol, and adaptively sets a slow-start threshold according to the available bandwidth. The method includes initializing a congestion window 'cwnd,' sending cwnd packets, estimating an available bandwidth for the cwnd packets. The congestion window cwnd is set to a higher number, and the higher number of further packets is sent if the available bandwidth is greater than a first threshold level. The available bandwidth is re-estimated for the higher number of the further packets, and a soft start threshold 'ssthresh' is set to the re-estimated available bandwidth. A statistical measure is calculated for the re-estimated available bandwidth, and the congestion window cwnd is set equal to ssthresh if a ratio of the statistical measure to the re-estimated available bandwidth is less than a second threshold level.

System and method for transmission control protocol slow-start

Technologies are presented for power optimization of datacenter networks in a hierarchical perspective. In some examples, a two-level power optimization model may be established to reduce the power consumption of datacenter networks by switching off network switches and links while still guaranteeing full connectivity and maximum link utilization. The model may be implemented by solving a capacitated constraint multi-commodity flow (CMCF) problem employing simple heuristic techniques. A power status of network switches may be determined according to a network traffic matrix and the CMCF optimization determined at core-level and at pod-level. A complementary process to provision whole network connectivity and to meet quality of service (QoS) goals may also be performed.

Growing community of inventors

Harrison, NJ, United States of America

Yan Zhang