Hirochika Asai

Poptrie: A Compressed Trie with Population Count for Fast and Scalable Software IP Routing Table Lookup

Internet of Things leads to routing table explosion. An inexpensive approach for IP routing table lookup is required against ever growing size of the Internet. We contribute by a fast and scalable software routing lookup algorithm based on a multiway trie, called Poptrie. Named after our approach to traversing the tree, it leverages the population count instruction on bit-vector indices for the descendant nodes to compress the data structure within the CPU cache. Poptrie outperforms the state-of-the-art technologies, Tree BitMap, DXR and SAIL, in all of the evaluations using random and real destination queries on 35 routing tables, including the real global tier-1 ISPs full-route routing table. Poptrie peaks between 174 and over 240 Million lookups per second (Mlps) with a single core and tables with 500--800k routes, consistently 4--578% faster than all competing algorithms in all the tests we ran. We provide the comprehensive performance evaluation, remarkably with the CPU cycle analysis. This paper shows the suitability of Poptrie in the future Internet including IPv6, where a larger route table is expected with longer prefixes.

Tracking the Evolution and Diversity in Network Usage of Smartphones

We analyze the evolution of smartphone usage from a dataset obtained from three, 15-day-long, user-side, measurements with over 1500 recruited smartphone users in the Greater Tokyo area from 2013 to 2015. This dataset shows users across a diverse range of networks; cellular access (3G to LTE), WiFi access (2.4 to 5GHz), deployment of more public WiFi access points (APs), as they use diverse applications such as video, file synchronization, and major software updates. Our analysis shows that smartphone users select appropriate network interfaces taking into account the deployment of emerging technologies, their bandwidth demand, and their economic constraints. Thus, users show diversity in both how much traffic they send, as well as on what networks they send it. We show that users are gradually but steadily adopting WiFi at home, in offices, and public spaces over these three years. The majority of light users have been shifting their traffic to WiFi. Heavy hitters acquire more bandwidth via WiFi, especially at home. The percentage of users explicitly turning off their WiFi interface during the day decreases from 50% to 40%. Our results highlight that the offloading environment has been improved during the three years, with more than 40% of WiFi users connecting to multiple WiFi APs in one day. WiFi offload at offices is still limited in our dataset due to a few accessible APs, but WiFi APs in public spaces have been an alternative to cellular access for users who request not only simple connectivity but also bandwidth-consuming applications such as video streaming and software updates.

Network application profiling with traffic causality graphs

SUMMARY A network application profiling framework is proposed that is based on traffic causality graphs (TCGs), representing temporal and spatial causality of flows to identify application programs. The proposed framework consists of three modules: the feature vector space construction using discriminative patterns extracted from TCGs by a graph-mining algorithm; a feature vector supervised learning procedure in the constructed vector space; and an application identification program using a similarity measure in the feature vector space. Accuracy of the proposed framework for application identification is evaluated, making use of ground truth packet traces from seven peer-to-peer (P2P) application programs. It is demonstrated that this framework achieves an overall 90.0% accuracy in application identification. Contributions are twofold: (1) using a graph-mining algorithm, the proposed framework enables automatic extraction of discriminative patterns serving as identification features; 2) high accuracy in application identification is achieved, notably for P2P applications that are more difficult to identify because of their using random ports and potential communication encryption. Copyright

P2V Migration with Post-Copy Hot Cloning for Service Downtime Reduction

We propose a post-copy hot cloning mechanism to reduce service downtime of physical to virtual (P2V) migration. The conventional P2V migration requires to copy disk images from a physical to a virtual machine with stopping the system to ensure the consistency of the disk images, but it usually takes long time (e.g., several hours) because of slow I/O performance of storage devices. The proposed hot cloning mechanism copies disk images from a physical machine to a virtual machine in the background with running the system on the virtual machine. Since the disk image copy is the bottleneck of P2V migration, this hot cloning mechanism drastically reduces the service downtime. To ensure the consistency of disk images during the hot cloning, the hot cloning mechanism provides a block device of the source physical machine and a disk image (or block device) of the destination virtual machine as one virtualized block device that manages all I/O operations to the virtual machine. In this paper, we implement this post-copy hot cloning mechanism using the Linux Network Block Device (NBD) kernel module to support Linux-based hyper visors such as Xen and Linux KVM. We discuss the service downtime that is reduced by the proposed mechanism, and demonstrate that the implementation performs well in practical environments throughout the performance evaluations on the copy speed, hyper visor load, and system performance on a virtual machine.

Where are the bottlenecks in software packet processing and forwarding?: Towards high-performance network operating systems

We analyze the performance bottlenecks in packet processing and forwarding with general-purpose computers to achieve high-throughput and low-latency, and to guarantee service quality. In this paper, we first demonstrate that the performance bottleneck in software packet processing and forwarding is a blocking operating to access to PCIe registers with memory mapped I/O that are used to operate the head and tail pointers of ring buffers of NICs through two experiments. We then summarize the strategies to achieve high-performance software packet processing and forwarding against the bottlenecks, and develop and evaluate a prototype system with routing functionality to illustrate these strategies are effective.

Towards characterization of wireless traffic in coexisting 802.11a/g and 802.11n network

The number of wireless devices and the traffic volume generated by these devices become significant today, and many devices begin supporting 802.11n protocol for higher-speed wireless access. However, the diversification in link types of end-hosts may degrade system performance. For example, hosts using 802.11 protocols had better not be relay nodes in a P2P live streaming system because 802.11 is a half-duplex protocol and usually less stable compared to modern wired links. Hence, understanding traffic characteristics of various link types is essential for improving or building network architectures. Moreover, estimation of the link type of a remote host possibly achieves better performance (e.g., higher throughput) in some network systems. Baiamonte et al. [1] have proposed an algorithm to detect wireless hosts from passive measurement by using the entropy of packet interarrival time (PIT). Wei et al. [2] have also proposed an algorithm to classify access network types. However, these algorithms pay no attention to new link types such as 802.11n, 3G, and WiMAX even though each of them has different characteristics and possibly degrades the performance of network systems.

Spatio-Temporal Modeling of BGP Routing Table Evolution

Modeling the routing table growth is vital to the BGP operation. The temporal evolution of the routing table size has been researched to anticipate the memory limitation of BGP routers. However, route aggregation and compression techniques make it difficult to expect the actual memory size from the routing table size. Therefore, further evolution models focusing on the spatial routing table structure are required to evaluate the tolerance of these techniques to future routing table growth. In this paper, we create an evolution model focusing on spatio-temporal route changes using ten-year BGP routing table datasets. We categorize route variations into three types; new, fragmented, and vanished and then create a model for each type. We also demonstrate the characteristics of the model parameters in these ten years.

Towards interdomain transit traffic reduction in peer-assisted content delivery networks

Transit traffic exchanged with transit providers costs more for network providers compared to intra-domain traffic or traffic exchanged over peering links. We have measured and analyzed the peer distribution in BitTorrent, which is one of peer-assisted content delivery networks (CDNs). From the peer distribution, we show the potential of the high-cost transit traffic reduction. We then propose a peer selection preference which takes into account the economical relationships among Autonomous Systems (ASes) in peer-assisted CDNs to reduce the high-cost transit traffic. Since most commercial Internet service providers cannot disclose the relationships due to their commercial contract, we employ degree-based heuristics for inferring the relationships; degree can be approximated from publicly available BGP routing tables. We show that the peer selection method utilizing the proposed preference can reduce interdomain transit traffic exchanged with provider ASes by trace-driven computer simulation. The significance of this paper are 1) we show the potential of the transit traffic reduction from peer distribution analysis, and 2) the peer selection method with the proposed preference appropriately reduces the high-cost transit traffic with degree-based AS relationships inference heuristics even though there is no public AS relationships information.