Network function virtualization (NFV) enables network operators to flexibly provide diverse virtualized functions for services such as Internet of things (IoT) and mobile applications. To meet multiple quality of service (QoS) requirements against time-varying network environments, infrastructure providers must dynamically adjust the amount of computational resources, such as CPU, assigned to virtual network functions (VNFs). To provide agile resource control and adaptiveness, predicting the virtual server load via machine learning technologies is an effective approach to the proactive control of network systems. In this paper, we propose an adjustment mechanism for regressors based on forgetting and dynamic ensemble executed in a shorter time than that of our previous work. The framework includes a reducing training data method based on sparse model regression. By making a short list of training data derived from the sparse regression model, the relearning time can be reduced to about 57% without degrading provisioning accuracy.

Advanced information-processing services based on cloud computing are in great demand. However, users want to be able to customize cloud services for their own purposes. To provide image-processing services that can be optimized for the purpose of each user, we propose a technique for chaining image-processing functions in a CPU-field programmable gate array (FPGA) coupled server architecture. One of the most important requirements for combining multiple image-processing functions on a network, is low latency in server nodes. However, large delay occurs in the conventional CPU-FPGA architecture due to the overheads of packet reordering for ensuring the correctness of image processing and data transfer between the CPU and FPGA at the application level. This paper presents a CPU-FPGA server architecture with a real-time packet reordering circuit for low-latency image processing. In order to confirm the efficiency of our idea, we evaluated the latency of histogram of oriented gradients (HOG) feature calculation as an offloaded image-processing function. The results show that the latency is about 26 times lower than that of the conventional CPU-FPGA architecture. Moreover, the throughput decreased by less than 3.7% under the worst-case condition where 90 percent of the packets are randomly swapped at a 40-Gbps input rate. Finally, we demonstrated that a real-time video monitoring service can be provided by combining image processing functions using our architecture.

Many studies have revealed that the performance of software-based Virtual Network Functions (VNFs) is insufficient for mission-critical networks. Scaling-out approaches, such as auto-scaling of VNFs, could handle a huge amount of traffic; however, the exponential traffic growth confronts us the limitations of both expandability of physical resources and complexity of their management. In this paper, we propose a fast datapath processing method called Packet Aggregation Flow (PA-Flow) that is based on hop-by-hop packet aggregation for more efficient Service Function Chaining (SFC). PA-Flow extends a notion of existing intra-node packet aggregation toward network-wide packet aggregation, and we introduce following three novel features. First, packet I/O overheads at intermediate network devices including NFV-nodes are mitigated by reduction of packet amount. Second, aggregated packets are further aggregated as going through the service chain in a hop-by-hop manner. Finally, next-hop aware packet aggregation is realized using OpenFlow-based flow tables. PA-Flow is designed to be available with various VNF forms (e.g. VM/container/baremetal-based) and virtual I/O technologies (e.g. vhost-user/SR-IOV), and its implementation does not bring noticeable delay for aggregation. We conducted two evaluations: (i) a baseline evaluation for understanding fundamental performance characteristics of PA-Flow (ii) a simulation-based SFC evaluation for proving PA-Flow's effect in a realistic environment. The results showed that throughput of short packet forwarding was improved by 4 times. Moreover, the total number of packets was reduced by 93% in a large-scale SFC.

The heart of Network Functions Virtualization (NFV) is both the softwarization of existing network middleboxes as Virtual Network Functions (VNFs) and the Service Function Chaining (SFC), also known as Service Chaining of them. Most existing VNFs are realized as VM-based general purpose appliances and shared by multiple user VMs. However, the cover range of VNF can be extended to directly reinforce network functionality of user VMs by introducing VM-specific VNFs. In this study, we propose micro-VNFs (µVNFs) and a VM-specific service chaining framework (vNFChain). Micro-VNFs are VM-specific lightweight VNFs that directly attach to a user VM, and can support not only traditional L2-L4 protocols but also stateful custom L7 protocols. The vNFChain framework constructs local service chains of µVNFs and transparently attaches the chain to the VM. Importantly, our framework achieves zero touch configuration for user VMs as well as no modification for existing system environments, such as virtual switch, hypervisor, and OS. In this paper, we describe architectural design and implementation of the framework. In addition, we evaluate the proposed approach in terms of throughput and CPU usage by comparing it with a DPDK-enabled VM-based µVNF model.