Efficiently scheduling data processing jobs on distributed compute clusters requires complex algorithms. Current systems use simple, generalized heuristics and ignore workload characteristics, since developing tuning a policy for each is infeasible. In this paper, we show that modern machine learning techniques can generate highly-efficient policies automatically.

While many deep learning (DL)-based networking systems have demonstrated superior performance, the underlying Deep Neural Networks (DNNs) remain blackboxes and stay uninterpretable for network operators. The lack of interpretability makes DL-based prohibitive to deploy in practice. In this paper, we propose Metis, a framework that provides two general categories problems spanning local global control. Accordingly, Metis introduces different interpretation methods based on decision tree...

Real-time communication (RTC) applications like video conferencing or cloud gaming require consistent low latency to provide a seamless interactive experience. However, wireless networks including WiFi and cellular, albeit providing satisfactory median latency, drastically degrade at the tail due frequent substantial bandwidth fluctuations. We observe that control loop for sending rate of RTC is inflated when congestion happens access point (AP), resulting in untimely adaption dynamics....

In network function virtualization (NFV), functions (NFs) are chained as a service chain (SFC) to enhance NF management with high flexibility. Recent solutions indicate that the processing performance of SFCs can be significantly improved by offloading NFs programmable switches. However, such requires deep understanding properties achieve maximum SFC performance, which brings non-trivial burdens administrators. this paper, we propose LightNF, novel system simplifies in networks. LightNF...

Network function virtualization (NFV) together with software defined networking (SDN) offers the potential for enhancing service delivery flexibility and reducing overall costs. Based on capability of dynamic creation destruction network (NF) instances, NFV provides great elasticity in NF control, such as scaling out, in, load balancing. To realize traffic flows need to be redistributed across instances. However, deciding which are suitable migration is a critical problem efficient control....

Major commercial client-side video players employ adaptive bitrate (ABR) algorithms to improve user quality of experience (QoE). With the evolvement ABR algorithms, increasingly complex methods such as neural networks have been adopted pursue better performance. However, these are too heavyweight be directly implemented in client devices, especially mobile phones with very limited resources. Existing solutions suffer from a trade-off between algorithm performance and deployment overhead. To...

Advanced network management systems, including measurement and traffic control, rely on a remote controller to make control decisions. However, this approach incurs long loop of few seconds minutes. Even if we switch switch-local controller, the latency is still tens milliseconds unacceptable for many latency-sensitive tasks. In paper, propose Martini, general framework that supports measurement-based timely control. The key idea perform measurement, decision, entirely in data plane. This...

In network function virtualization (NFV), functions (NFs) are chained as a service chain (SFC) to enhance NF management with low cost and high flexibility. Recent NFV solutions indicate that the packet processing performance of SFCs can be significantly improved by offloading NFs programmable switches. However, such requires deep understanding heterogeneous properties (e.g., resource consumption behaviors) achieve maximum SFC performance. Unfortunately, none existing provide automatic...

Major commercial client-side video players employ adaptive bitrate (ABR) algorithms to improve the user quality of experience (QoE). With evolvement ABR algorithms, increasingly complex methods such as neural networks have been adopted pursue better performance. However, these are too heavyweight be directly deployed in client devices with limited resources, mobile phones. Existing solutions suffer from a trade-off between algorithm performance and deployment overhead. To make sophisticated...

Smart Network Interface Cards (SmartNICs) have been widely used to accelerate software-based network functions (NFs). However, from the scope of a service chain, careless selection NFs offload onto SmartNIC could severely degrade performance due frequent communications between CPU and SmartNIC. In this paper, we present SmartChain, high efficient framework that achieves optimal partition chains CPU. SmartChain consists two logical steps. First, analyzes suitability elements in chain run on...

Efficiently scheduling data processing jobs on distributed compute clusters requires complex algorithms. Current systems, however, use simple generalized heuristics and ignore workload characteristics, since developing tuning a policy for each is infeasible. In this paper, we show that modern machine learning techniques can generate highly-efficient policies automatically. Decima uses reinforcement (RL) neural networks to learn workload-specific algorithms without any human instruction...

The modularization of service function chains (SFCs) in network virtualization (NFV) could introduce significant performance overhead and resource efficiency degradation due to introducing frequent packet transfer consuming much more hardware resources. In response, we exploit the reusability, lightweightness, individual scalability features elements modularized SFCs (MSFCs) propose MicroNF, an efficient framework for MSFC NFV. MicroNF addresses problems three ways. First, graph constructor...

The modularization of Service Function Chains (SFCs) in Network Virtualization (NFV) could introduce significant performance overhead and resource efficiency degradation due to introducing frequent packet transfer consuming much more hardware resources. In response, we exploit the lightweight individually scalable features elements Modularized SFCs (MSFCs) propose CoCo, a compact optimized consolidation framework for MSFC NFV. CoCo addresses above problems two ways. First, Optimized Placer...

Network Function Virtualization (NFV) has the potential to enhance service delivery flexibility and reduce overall costs by provisioning software-based function chains (SFCs) on commodity hardware. However, we observe that existing CPU-based SFC solutions cannot achieve both high performance elasticity simultaneously. To address such a critical challenge, seek beyond CPU exploit capability of Graphics Processing Unit (GPU) support NFV. We propose GEN, GPU-based elastic framework for As...

Network Function Virtualization (NFV) together with Software Defined Networking (SDN) offers the potential for enhancing service delivery flexibility and reducing overall costs. Based on capability of dynamic creation destruction network function (NF) instances, NFV provides great elasticity in NF control, such as scaling out, in, load balancing, etc. To realize traffic flows need to be redistributed across instances. However, deciding which are suitable migration is a critical problem...

To improve the performance of mobile web service, a new transport protocol, QUIC, has been recently proposed. However, for large-scale real-world deployments, deciding whether and when to use QUIC in service is challenging. Complex temporal correlation network conditions, high spatial heterogeneity users nationwide deployment, limited resources on devices all affect selection protocols. In this paper, we present WiseTrans adaptively switch protocols online completion time requests.

To improve the performance of mobile web services, a new transport protocol, QUIC, has been recently proposed as substitute for TCP. However, with pros and cons it is challenging to decide whether when use QUIC in large-scale real-world services. Complex temporal correlation network conditions, high user heterogeneity nationwide deployment, implementation diversity variants limited, resources on devices all affect selection protocols. In this paper, we present WiseTrans, an adaptive protocol...

BGP is the only inter-domain routing protocol that plays an important role on Internet. However, suffers from route leak, which can cause serious security threats. To mitigate effects of accurate and timely leak location great importance. Prior studies leverage AS business relationships to locate in real time. they fail achieve high accuracy. Recent apply machine learning accurately detect statistical features massive messages. Nevertheless, have detection latency cannot further leak. In...

Network Function Virtualization (NFV) provides middleboxes with substantial elasticity from a system level, and Artificial Neural (ANN) empowers great intelligence an algorithm-level perspective. However, when ANN-based Functions (NFs) want to take advantage of the NFV, our study finds that huge gaps exist between existing approaches ideal goals for control NFs. By revealing key differences NFs traditional NFs, we propose LEGO, innovative framework systematic mechanisms traffic splitting,...

With a high frame-rate and bit-rate, ultra-high definition (UHD) real-time communication (RTC) users could sometimes suffer from severe service degradation. Due to the fluctuations of frames incoming decoding at client side, decoder queue be formulated before streaming side. Those easily overload introduce noticeable delay for those queued frames. In this paper, we propose Frame-Skipping mechanism effectively reduce queuing by actively managing inside queue. We jointly optimize with skipping...

Real-time applications are tremendously emerging in recent years. From video conferencing to virtual reality and quantitative trading, real-time require an extremely low latency of < 100 ms or even lower [8]. Operators for decades have deployed content delivery networks, edge access nodes numerous architectures reduce the RTT be as 10--20 [11]. Also, some (e.g., WebRTC [1]) choose rely on UDP sockets since offers more flexibility loss recovery congestion control.

Since the beginning of decade, real-time communication (RTC) protocols, including video conferencing applications, cloud gaming, and VR/AR streaming, have rapidly expanded over Internet. With eventual arrival metaverse, RTC traffic is widely anticipated to become predominant type Internet in near future [4, 9].