Today's high-speed switches employ an on-chip shared packet buffer. The buffer is becoming increasingly insufficient as it cannot scale with the growing switching capacity. Nonetheless, needs to face highly intense bursts and meet stringent performance requirements for datacenter applications. This imposes rigorous demand on Buffer Management (BM) scheme, which dynamically allocates across queues. However, de facto BM designed over two decades ago, ill-suited of today's network. In this...

In data center networks, micro-burst is a common traffic pattern and the packet dropping caused by it usually leads to serious performance degradation. Meanwhile, most of current commodity switches employ on-chip shared memory, buffer management policies them ensure fair sharing memory among all ports. Among various polices, Dynamic Threshold (DT) widely used switch vendors. However, because DT needs reserve fraction buffer, there free space while packets from are dropped. this paper, we...

Due to the explosive growth of all kinds Internet services, data centers have become an irreplaceable and vital infrastructure support this soaring trend. Compared with traditional networks, center networks (DCNs) unique features, such as high bandwidth, low latency, many-to-one communication mode, shallow buffered switches, multi-root topology. These new characteristics pose a lot challenges previous network technics (e.g., Ethernet, Equal Cost Multi-path (ECMP), TCP), making them hard...

Today's data center switches usually employ on-chip shared memory; buffer management policy in them is essential to ensure fair sharing of memory among all ports. Among various polices, Dynamic Threshold (DT) widely used by switch vendors. Meanwhile, centers, distributed applications such as MapReduce often introduce micro-burst traffic into network and the packet dropping caused leads serious performance degradation. When arrives at switches, DT unable fully utilize absorb it. Therefore,...

Currently, Ethernet is being enhanced to become the unified switch fabric in data centers. With fabric, cost on redundant devices reduced, while design and management of center networks are simplified. Congestion one indispensable enhancements Ethernet, Quantized Notification (QCN) has just been ratified as formal standard. Though QCN investigated for several years, there exist few in-depth theoretical analyses QCN. The most possible reason that heuristically designed involves property...

Most commercial players adopt adaptive bitrate (ABR) algorithms to dynamically decide each chunk's based on the perceived network bandwidth and buffer occupancy. However, current ABR are agnostic of audio selection since they deem it has negligible influence video due small size chunks. Nevertheless, with development technologies, content increases dramatically in recent years. Thus, inappropriate can significantly affect deteriorate viewing experience. To tackle these inefficiencies, we...

Mobile video streaming dominates today's network traffic, and adaptive bitrate (ABR) algorithms have been routinely adopted for transmitting media content across dynamic mobile networks. State-of-the-art ABR mainly alter without considering audio as they consider the impact on negligible due to their small size. However, bring users an immersive experience, recent providers applied high-quality with large sizes, like stereophonic sound. Therefore, improper selection will adversely affect...

Copa is a delay-based congestion control algorithm proposed in NSDI recently. It can achieve consistent high performance under various network environments and has already been deployed Facebook. In this paper, we theoretically analyze reveal its large queuing delay poor fairness issue certain conditions. The root cause that fails to clear the bottleneck buffer occupancy periodically as expected. Accordingly, may get wrong base RTT estimation enter competitive mode by mistake, leading...

Traffic policing is widely used by ISPs to limit their customers' traffic rates. It has long been believed that a well-tuned policer offers satisfactory performance for TCP. However, we find this belief breaks with the emergence of new congestion control (CC) algorithms: flows using CC algorithms can easily occupy majority bandwidth, starving traditional TCP flows. We confirm problem experiments and reveal its root cause as follows. Without buffer in policers, only causes packet losses,...

Abstract Nowadays, the paradigm of mobile computing is evolving from a centralized cloud model towards Mobile Edge Computing (MEC). In regions without ground communication infrastructure, incorporating aerial edge nodes into network emerges as an efficient approach to deliver Artificial Intelligence (AI) services Ground Devices (GDs). The computation offloading and resource allocation problem within HAP-assisted MEC system investigated in this paper. Our goal minimize energy consumption....

The emerging latency-sensitive services such as smart grid and tactile internet require deterministic network performance including end-to-end latency, latency jitter bounded packet loss rate. To empower standard Ethernet with capability, we provide a forwarding system named DLCC congestion control queue management available at common hardwares to delay over Internet. Meanwhile, provides per-hop correction scheme minimize variation under dynamic scenarios. Compared other schemes, effectively...

Recently, Ethernet is being enhanced as the unified switch fabric of data centers, called Data Center Ethernet. The end-to-end congestion management one indispensable enhancements, and Quantized Congestion Notification (QCN) has been ratified to be standard. Our experiments show that QCN suffers from oscillation queue at bottleneck link. With changes system parameters network configurations, may become so serious emptied frequently. As a result, utilization link degrades. Theoretical...

Recently, the energy consumption of Ethernet has become one hottest topics focused by both academic committee and industry, especially with increase link speed from 1Gbps to 10Gbps nowadays or even 40/100/200Gbps in near future. To save consumed Ethernet, Energy Efficient (EEE) is developed standardized IEEE 802.3az work group. When there no incoming traffic, EEE can saves 90 % its entering into Low Power Idle (LPI) mode. maximize saving Burst TRansmission (BTR) algorithm, which defines a...

Memcached is a widely used in-memory caching solution in large-scale searching scenarios. The most pivotal performance metric latency, which affected by various factors including the workload pattern, service rate, unbalanced load distribution and cache miss ratio. To quantitate impact of each factor on we establish theoretical model for system. Specially, formulate among servers set probabilities, capture burst concurrent key arrivals at form batching blocks, add processing stage. Based...

At present, companies and standards organizations are enhancing Ethernet as the unified switch fabric for all of TCP/IP traffic, storage traffic high performance computing in data centers. Backward congestion notification (BCN) is basic mechanism end-to-end management enhancement Ethernet. To fulfill special requirements fabric, i.e., losslessness low transmission delay, BCN should hold buffer occupancy around a target point tightly. Thus, stability control loop size critical to BCN....

In current large-scale distributed key-value stores for cloud computing, the tail latency of hundreds accesses generated by an end-user request determines response time this request. Replica selection algorithms, which select best replica server each access as much possible, is crucial to reduce latency. This paper summarizes algorithms and classifies them into three categories: information-agnostic, client-independence feedback, according their demanded information. Furthermore,...

Recently, Ethernet is enhanced as the unified switch fabric of data centers, called center Ethernet. One indispensable enhancements end-to-end congestion management, and currently quantized notification (QCN) has been ratified corresponding standard. However, our experiments show that QCN suffers from large oscillations queue length at bottleneck link such buffer emptied frequently accordingly utilization degrades, with certain system parameters network configurations. This phenomenon to...

Data centers traffic is composed by numerous latency-sensitive "mice" flows, which consisted of only several packets, and a few throughput-sensitive "elephant" occupy more than 80% overall load. Generally, the short-lived flows induce transient congestion long-lived cause persistent congestion. The network major performance inhibitor. Conventionally, hop-by-hop end-to-end flow control mechanisms are employed to relief congestion, respectively. However, in face mixture elephants mice, we find...