Non-Volatile Memory Express (NVMe) over TCP is an efficient technology for accessing remote Solid State Drives (SSDs); however, it may cause a serious interference issue when used in a containerized environment. In this study, we propose a CPU isolation scheme for NVMe over TCP in such an environment. The proposed scheme measures the CPU usage of the NVMe over TCP worker, charges it to containers in proportion to their NVMe traffic, and schedules containers to ensure isolated sharing of the CPU. However, because the worker runs with a higher priority than normal containers, it may not be possible to achieve CPU isolation with container scheduling alone. To solve this problem, we also control the CPU usage of the worker by throttling NVMe over TCP traffic. The proposed scheme is implemented on a real testbed for evaluation. We perform extensive experiments with various workloads and demonstrate that the scheme can provide CPU isolation even in the presence of excessive NVMe traffic.

Despite the advent of various network enhancement technologies, it is yet a challenge to provide high-performance networking for GPU-accelerated applications on commodity Ethernet. Kernel-bypass I/O, such as DPDK or netmap, which is normally optimized for host memory-based CPU applications, has limitations on improving the performance of GPU-accelerated applications due to the data transfer overhead between host and GPU. In this paper, we propose GPU-Ether, GPU-native packet I/O on commodity Ethernet, which enables direct network access from GPU via dedicated persistent kernel threads. We implement GPU-Ether prototype on a commodity Ethernet NIC and perform extensive testing to evaluate it. The results show that GPU-Ether can provide high throughput and low latency for GPU applications.

In Information-Centric Networking (ICN), in-network caching is one of the core functions to implement efficient content distribution. As content is served from the network cache, redundant transmission across the same link is reduced and better quality services are provided to the user. However, caching operations (locating cached content, validating/writing content to the cache) may be a large burden on routers. In this article, we substantially reduce the caching overhead of routers by introducing T-Caching, a new in-network caching model in which routers selectively cache some of the most popular content recommended by content providers. In T-Caching, tokens are used 1) to enable routers to maximize caching benefits by simply controlling the amount of content inserted into the cache; 2) to allow content providers to recommend the most popular content as much as routers would actually cache. Our simulation and empirical studies using synthetic data, as well as real-world traces, show that T-Caching allows routers to insert much less than 1 percent of the content into the cache, compared to typical caching mechanisms. Nevertheless, the cache-hit ratio is improved by up to 2~9 times depending on the cache size. We believe that T-Caching can significantly contribute to improving feasibility and performance of in-network caching in ICN.

In Named-Data Networking (NDN), content is cached in network nodes and served for future requests. This property of NDN allows attackers to inject poisoned content into the network and isolate users from valid content sources. Since a digital signature is embedded in every piece of content in NDN architecture, poisoned content is discarded if routers perform signature verification; however, if every content is verified by every router, it would be overly expensive to do. In our preliminary work, we have suggested a content verification scheme that minimizes unnecessary verification and favors already verified content in the content store, which reduces the verification overhead by as much as 90% without failing to detect every piece of poisoned content. Under this scheme, however, routers are vulnerable to verification attack, in which a large amount of unverified content is accessed to exhaust system resources. In this paper, we carefully look at the possible concerns of our preliminary work, including verification attack, and present a simple but effective solution. The proposed solution mitigates the weakness of our preliminary work and allows this paper to be deployed for real-world applications.

This paper investigates the impact of large flows in Content Centric Networks (CCN). In CCN, routers have caches, and store all the data after forwarding. If large flows temporarily occupy a content cache, they may evict popular chunks from the cache, and it results in low cache hit ratio. We mathematically analyzed the amount of occupancy of a large flow in a cache, and realized that a few large flows can constitute a significant portion of a cache. Our simulation results showed that small flows experience cache hit ratio degradation as the number of large flows increases. Finally, we present that limiting the occupancy of large flows in a cache can effectively improve the cache hit ratios for small flows.

Two major goals of queue management are flow fairness and queue-length stability However, most prior works dealt with these goals independently. In this paper, we show that both goals can be effectively achieved at the same time. We propose a novel scheme that realizes flow fairness and queue-length stability. In the proposed scheme, high-bandwidth flows are identified via a multilevel caching technique. Then, we calculate the base drop probability for resolving congestion with a stable queue, and apply it to individual flows differently depending on their sending rates. Via extensive simulations, we show that the proposed scheme effectively realizes flow fairness between unresponsive and TCP flows, and among heterogeneous TCP flows, while maintaining a stable queue.

In this paper, we propose a novel technique to deal with sudden bandwidth changes in TCP. In the current Internet, sudden bandwidth changes may occur due to vertical handovers between heterogeneous access networks or routing path changes. The current implementation of TCP is designed and optimized for stable networks, and does not adapt well upon sudden bandwidth changes. Consequently, it might suffer from packet losses in burst upon sudden decrements of bandwidth and under-utilization upon sudden increments of bandwidth. To resolve this problem, we propose to modify the current TCP algorithm to include a new phase, called fast adaptation (FA). The FA phase is triggered upon detecting sudden bandwidth changes, and a TCP sender in the FA phase attempts to recover lost packets quickly to avoid timeout upon sudden decrement of bandwidth. Upon sudden increment of bandwidth, it increases its window size drastically to realize full-utilization. Through extensive simulations and experiments, it is shown that the proposed scheme can effectively deal with sudden bandwidth changes.

The Assured Forwarding (AF) service in a differentiated services network offers different levels of forwarding assurance for IP packets. Several studies have proposed to manipulate the contention window size to provide service differentiation in wireless networks. However, since network congestion in multi-hop wireless networks causes packet losses at the link layer, the contention window size based differentiation scheme is not sufficient to provide the AF service. In this paper, we propose a rate control scheme to control load on the shared channel by adjusting sending rate according to the estimated channel state and the amount of AF traffic. We further propose a RTS retry limit adaptation scheme to compensate the rate control scheme when the accurate estimation of the channel state is not easy. Throughout extensive simulations, we show that proposed schemes are effective to differentiate forwarding assurance in multi-hop wireless networks.

We propose a fair bandwidth allocation scheme for multi-radio multi-channel wireless mesh networks (WMNs) using distributed algorithm. Through an extensive simulation, we show that our scheme ensures per node fairness without loss of the total aggregate throughput.

In a mobile ad hoc network (MANET), it has been addressed that packet losses due to collision are often misinterpreted as routing failures, and cause unnecessary overhead for routing maintenance. There have been several attempts to avoid the unnecessary overhead through reducing collision losses. They are effective in a static topology where most losses are due to collision. In a dynamic topology, however, packets are lost due to actual routing failures (induced by mobility) as well as due to collision, and efforts for reducing collision are not enough. In this paper, we propose a new scheme for adjusting the limit of RTS retransmissions. In the proposed scheme, we treat packet losses differently as follows: (a) upon collisions, we increase the limit to reduce collision losses; and (b) upon routing failures, we decrease the limit to avoid unnecessary retransmissions. Through extensive simulations, it is shown that the proposed scheme effectively improves throughput in various scenarios and outperforms other comparable schemes.

Queue length oscillation at a congested link causes many undesirable properties such as large delay jitter, underutilization of the link and packet drops in burst. The main reason of this oscillation is that most queue management schemes determine the drop probability based on the current traffic without consideration on the impact of that drop probability on the future traffic. In this paper, we propose a new active queue (AQM) scheme to reduce queue oscillation and realize stable queue length. The proposed scheme measures the current arrival and drop rates, and uses them to estimate the next arrival rate. Based on this estimation, the scheme calculates the drop probability which is expected to realize stable queue length. We present extensive simulation with various topologies and offered traffic to evaluate performance of the proposed scheme. The results show that the proposed scheme remarkably reduces queue length oscillation compared to other well-known AQMs. It is also shown that the proposed scheme improves fairness among TCP flows due to the stable drop probability, and maintains high utilization with small queue length.

Regarding to address allocation, a mobile ad hoc network (MANET) may suffer from the lack of address and network partition/merging due to mobility of nodes. In this paper, we propose a new address allocation protocol for dealing with those problems. The proposed protocol is developed based on disjoint address set distribution with binary splitting for scalability, and provides special treatments for resolving the lack of addresses. We also present an effective technique for handling network partition and merging. Through simulation, we show that the proposed protocol is effective to allocate addresses in a MANET with reasonable latency and communication overhead.

As more wireless access networks become available, there is a new challenge emerged: how to discover currently available radio access networks (RANs). To discover available RANs without help from centralized servers, each mobile host periodically turns on network interfaces to discover available networks. Performance of a RAN discovery scheme may depend on how to maintain a valid list of available networks while minimizing power consumption to discover RANs. In this paper, we derive several decentralized RAN discovery schemes. Through extensive simulations, we evaluate them in terms of achieved bandwidth, the number of handovers and power consumption.

The differentiated services architecture is receiving wide attention as a framework for providing different levels of service according to a service profile in the Internet. The current architecture allows aggregation of flows sharing a service profile. This paper looks at the problem of achieving specific QoS goals of individual flows by flexibly managing resources available to an aggregated source. We derive a simple analytic model for the relationship between per-session behavior, aggregate packet marking and packet differentiation within a DiffServ network. The paper presents an adaptive marker based on a TCP performance model within a DiffServ network. The paper shows that an aggregated marker can maintain the state of individual flows at the edge of the network and utilize this state effectively in adaptively marking packets of individual flows to meet their QoS goals.

The differentiated services architecture has been proposed for providing different levels of services and has received wide attention. A packet in a diff-serv domain is classified into a class of service according to its contract profile and treated differently by its class. While many studies have addressed issues on the diff-serv architecture (e.g., dropper, marker, classifier and shaper), there have been few attempts to analytically understand a flow's behavior in a diff-serv network. We propose simple models of TCP behavior in a diff-serv network. Our models quantitatively characterize TCP throughput as functions of the contract rate, the packet-drop rate and the round-trip time in either two-drop precedence or three-drop precedence network. We also extend our models to aggregated flows. The models are validated through a number of simulations.