Low latency in establishing a communication link is a key to maximize the performance of vehicle-to-vehicle wireless LANs (V2V WLANs), as contact duration of vehicles is often limited due to their fast mobility. However, previous works have mainly addressed Vehicle-to-Infrastructure (V2I) link setup latency on legacy Wi-Fi standards (IEEE 802.11b/g). Moreover, there has been a lack of performance evaluation regarding V2V link setup latency with current Wi-Fi technology and IEEE 802.11ai standard defines Fast Initial Link Setup (FILS), which simplifies the link setup process to reduce the latency. In this paper, we investigate the characteristics of V2V link setup latency with regular Extensible Authentication Protocol (EAP) authentication (EAP-TLS) and FILS with current Wi-Fi technology. The evaluation was conducted in general vehicular scenarios with different combinations of inter-vehicle distance, vehicle speed and type of roads (i.e., urban roads vs highways). FILS achieved the average link setup latency of 0.32 seconds, outperforming EAP-TLS that resulted in the average latency of 2.35 seconds. The link setup latency was stable in most of the scenarios we tested regardless of inter-vehicle distance, speed and Wi-Fi signal strength. The only exception was on highways, where significantly longer link setup latency was observed when the relative speed and distance changed rapidly, line-of-sight (LoS) was fully blocked by other vehicles and there were no other objects (e.g., buildings and vegetation) that form indirect signal propagation paths.

Edge computing is becoming ever more relevant to offload compute-heavy tasks in vehicular networks. In this context, the concept of vehicular micro clouds (VMCs) has been proposed to use compute and storage resources on nearby vehicles to complete computational tasks. As many tasks in this application domain are time critical, offloading to the cloud is prohibitive. Additionally, task deadlines have to be dealt with. This paper addresses two main challenges. First, we present a task migration algorithm supporting deadlines in vehicular edge computing. The algorithm is following the earliest deadline first model but in presence of dynamic processing resources, i.e., vehicles joining and leaving a VMC. This task offloading is very sensitive to the mobility of vehicles in a VMC, i.e., the so-called dwell time a vehicles spends in the VMC. Thus, secondly, we propose a machine learning-based solution for dwell time prediction. Our dwell time prediction model uses a random forest approach to estimate how long a vehicle will stay in a VMC. Our approach is evaluated using mobility traces of an artificial simple intersection scenario as well as of real urban traffic in cities of Luxembourg and Nagoya. Our proposed approach is able to realize low-delay and low-failure task migration in dynamic vehicular conditions, advancing the state of the art in vehicular edge computing.

Decentralized content distribution over vehicle-to-vehicle (V2V) WLANs holds promise to mitigate load on cellular networks. The delay in establishing a communication link is a key to maximize V2V data transfer opportunity, as contact duration of vehicles is often limited due to their fast mobility. IEEE 802.11ai standard defines Fast Initial Link Setup (FILS), which simplifies the link setup process to reduce latency. In this paper, we investigate the benefit and limitations of FILS in V2V WLANs. We develop a testbed involving multiple vehicles, and compare the link setup latency of FILS with regular EAP authentication. Counterintuitively, regular EAP achieved equivalent performance in V2V WLANs due to architectural constraints of FILS.

We present the performance analysis of TCP CUBIC/BBR over V2V Wi-Fi (IEEE 802.11ac). Our measurements focus on three static parking scenarios with different distances at the office area. The results reveal the impact of TCP CUBIC and BBR on data transfer time and TCP metrics. (i) There are two major reasons of fluctuated TCP throughput. The first reason is narrow available bandwidth over V2V Wi-Fi. The second reason is delayed TCP connection establishment due to delayed SYN+ACK and SYN packet retransmission. (ii) The bytes in-flight of TCP CUBIC are dynamically changed by packet retransmission events on V2V Wi-Fi. The loss-based congestion control is not promising the high throughput. We believe that our analysis results provide implications for efficient data transfer over V2V Wi-Fi.

Vehicular Micro Cloud (VMC) is a group of connected vehicles where vehicles collaborate on a task over the vehicular network. A potential use case of VMC is that micro cloud members transfer data to each other via Vehicle-to-Vehicle (V2V) links, and the data is collaboratively uploaded to remote server (e.g., data center) when the connected vehicles are connected to a Wi-Fi network. In this paper, we focus on this use case and propose collaborative upload by VMC. We demonstrated the feasibility of the proposed method through the large-scale urban simulation (Los Angeles downtown traffic model). Our simulation results showed that the proposed method can reduce the upload data of traditional cellular network-based data upload by 50%.

Summary The Precision Time Protocol (PTP) is a widely used protocol for high‐precision time synchronization, but it requires hardware or driver support for network interface card. Therefore, we propose a portable time synchronization method that is independent of them. Our method uses eBPF to synchronize time by recording timestamps of PTP packets in the kernel. eBPF is provided by the kernel and is backward‐compatible, making it independent of hardware or other software. To demonstrate the portability and high‐precision time synchronization capabilities of our method, we compared the time synchronization precision in physical environment and virtual machine. We also modified server settings that could affect time synchronization precision, and clarified the effects of these changes. The experiments showed that our method can achieve the similar level of time synchronization precision as conventional methods while remaining portable and independent of other components.

Microservice is an architectural style widely adopted in various latency-sensitive cloud applications. Similar to the monolith, autoscaling has attracted the attention of operators for managing the resource utilization of microservices. However, it is still challenging to optimize resources in terms of latency service-level-objective (SLO) without human intervention. In this paper, we present GRAF, a graph neural network-based SLO-aware proactive resource autoscaling framework for minimizing total CPU resources while satisfying latency SLO. GRAF leverages front-end workload, distributed tracing data, and machine learning approaches to (a) observe/estimate the impact of traffic change (b) find optimal resource combinations (c) make proactive resource allocation. Experiments using various open-source benchmarks demonstrate that GRAF successfully targets latency SLO while saving up to 19% of total CPU resources compared to the fine-tuned autoscaler. GRAF also handles a traffic surge with 36% fewer resources while achieving up to 2.6x faster tail latency convergence compared to the Kubernetes autoscaler. Moreover, we verify the scalability of GRAF on large-scale deployments, where GRAF saves 21.6% and 25.4% for CPU resources and memory resources, respectively.

Congestion control algorithms using recurrent neural network (RNN) for bandwidth prediction are expected to improve throughput. Previous studies involving performance evaluations were conducted only using simulated data. However, simulation and real-world environments are largely different and rarely provide equivalent prediction accuracy. Therefore, we will verify whether our proposed method provides better prediction accuracy in a real-world environment. We measured communications in a real environment and generated training data by converting packet captured data with measurement of prediction accuracy on the generated data. The results showed that the maximum percentage of correct responses was 79.71%, which was comparable to the results obtained using simulated data.

Microservice is an architectural style that breaks down monolithic applications into smaller microservices and has been widely adopted by a variety of enterprises. Like the monolith, autoscaling has attracted the attention of operators in scaling microservices. However, most existing approaches of autoscaling do not consider microservice chain and severely degrade the performance of microservices when traffic surges. In this paper, we present pHPA, an autoscaling framework for the microservice chain. pHPA proactively allocates resources to the microservice chains and effectively handles traffic surges. Our evaluation using various open-source benchmarks shows that pHPA reduces 99%-tile latency and resource usage by up to 70% and 58% respectively compared to the most widely used autoscaler when traffic surges.

The software switches and routers are rapidly developed in the NFV/SDN area, and are expected to be adopted in the emerging future mobile network (beyond-5G). Despite the high expectation of software routers, its latency characteristics have not been investigated extensively. To provide the reference latency value for software-based network function, we analyzed the latency of SRv6 Mobile User Plane on VPP software router. We show that, 1) for the beyond-5G mobile network application, the latency is in the feasible range (roughly 3–30 μs), 2) the throughput is sufficient with the range in 1–5 Mpps per CPU core, 3) the latency has the strong correlation with the input traffic load, 4) and the latency can be stable in a carefully crafted implementation. The packet interval analysis shows the burstiness of its output traffic that is enforced by the processing of a vector (bulk/batch) of packets. Our latency characteristic analysis can be of help for improving the performance of software router using the batch processing and designing software-based beyond-5G mobile communication systems.

Microservice is an architectural style that has been widely adopted in various latency-sensitive applications. Similar to the monolith, autoscaling has attracted the attention of operators for managing resource utilization of microservices. However, it is still challenging to optimize resources in terms of latency service-level-objective (SLO) without human intervention. In this paper, we present GRAF, a graph neural network-based proactive resource allocation framework for minimizing total CPU resources while satisfying latency SLO. GRAF leverages front-end workload, distributed tracing data, and machine learning approaches to (a) observe/estimate impact of traffic change (b) find optimal resource combinations (c) make proactive resource allocation. Experiments using various open-source benchmarks demonstrate that GRAF successfully targets latency SLO while saving up to 19% of total CPU resources compared to the fine-tuned autoscaler. Moreover, GRAF handles traffic surge with 36% fewer resources while achieving up to 2.6x faster tail latency convergence compared to the Kubernetes autoscaler.

Due to the recent explosive increase in the use of wireless-LAN-connected terminals, such as notebook PCs and tablets, in enterprise environments, communication quality degradations caused by wireless LAN (WLAN) often occur. There are two reasons for this. The first reason is the bandwidth is narrow compared to that of wired LANs. The second one is the traffic volume at the WLAN access point (WLAN-AP) is abruptly changed by terminal-device movements. However, previous measurement methods cannot clarify that the degradation occurs at a WLAN section or which WLAN section is the cause. There are two kinds of methods to measure communication quality including that of WLANs: end-to-end throughput measurement methods and wireless-layer quality measurement methods. The former cannot detect the bottleneck point; the latter cannot detect the degradation of TCP-layer quality, which represents the QoE (quality of experience). To solve the issue mentioned above, we developed a measurement method that detects the degraded WLAN section that is the bottleneck and measures the TCP-layer quality via the bottleneck. Our passive measurement method analyzes the RTTs (round trip times) of TCP connections. Based on the nature of TCP that its RTT continuously increases due to a bottleneck WLAN section, we detect degraded TCP connections. By referring to the information of a detected TCP connection and connection information collected from WLAN-AP, we can detect the WLAN-AP in the bottleneck WLAN section. We implemented our method onto our WLAN-quality visualization tool. Using this tool, we also applied our method to an enterprise network with over a thousand hosts. In this paper, we refer to hosts when they contain both wired and wireless LAN terminals. WLAN terminals are called terminals. As a result, our method clearly detected the WLAN bottlenecks in the enterprise network, and we confirmed our method was useful to solve WLAN-related issues.

The GPRS Tunneling Protocol User Plane (GTP-U) has long been deployed for GSM, UMTS and 4G LTE. Now for 5G, IPv6 Segment Routing (SRv6) has been proposed as an alternative user plane protocol to GTP-U in both 3GPP and IETF. SRv6 based on source routing has many advantages: stateless traffic steering, network programming and so on. Despite the advantages, it is hard to expect to replace GTP-U by SRv6 all at once, even in a 5G deployment because of a lot of dependencies between 3GPP nodes. Therefore, stateless translation and coexistence with GTP-U have been proposed in IETF. However there are no suitable measurement platform and performance evaluation results between GTP-U and SRv6. In particular, it is hard to measure latency on commercial traffic generators when a receiving packet type is different from a sending packet type. In this paper, we focus on the performance evaluation between GTP-U and SRv6 stateless translation. We designed an SRv6 measurement platform using a programmable switch, and measured GTP-U and SRv6 functions with pre-defined scenarios on a local environment. Well-known performance metrics, such as throughput and packets per second (PPS), are measured by the traffic generator while the latency at the functions was measured using telemetry on our SRv6 platform. In our evaluation, we cannot find the abrupt performance drop of well-known metrics at SRv6 stateless translation. Moreover, the latency of SRv6 stateless translation is similar to GTP-U and their performance degradation is negligible. Through the evaluation results, it is obvious that the SRv6 stateless translation is acceptable to the 5G user plane.

Despite the growing popularity of enterprise virtual desktop infrastructure (VDI), little is known about its storage traffic characteristics. In addition, no prior work has considered the detailed characteristics of virtual machine (VM) behavior on VDI. In this paper, we analyze the enterprise storage traffic on commercial office VDI using designated VMs. For 28 consecutive days, we gathered various types of traces, including a usage questionnaire and active and passive measurements. To characterize the storage traffic, we focused on two perspectives: fibre channel (FC) traffic and VM behavior. From the FC traffic perspective, we found that read traffic is dominant, although the applications are similar to those in a previous small-scale VDI. In particular, the write response time of large transactions, e.g.,128 KiB, is strongly affected by a slight decrease in cache hits during an update storm. From the VM behavior, we found that all active user VMs generate only 25% of traffic. Although a few VMs generate massive traffic, their impact is small. These characteristics are unique in comparison with the small-scale VDI. Our results have significant implications for designing the next generation of VDI and improving its performance.

A scalable network with high bisection bandwidth and high availability requires efficient use of the multiple paths between pairs of end hosts. OpenFlow is an innovative technology and enables fine-grained, flow level control of Ethernet switching. However, the flow table structure defined by OpenFlow is not hardware friendly and the scalability is limited by the switch device. OpenFlow is also not sufficient for fast multipath failover. To overcome these limitations, we propose DomainFlow in which the network is split into sections and exact matches are used where possible to enable practical flow management using OpenFlow for commodity switches. We applied a prototype of DomainFlow to multipath flow management in the Virtual eXtensible LAN (VXLAN) overlay network environment. The total number of flow entries was reduced to 1/128 using currently available commodity switches, which was not possible before.