10th ACM Conference on Information-Centric Networking (ICN 2023)

Abstract:

Information-centric networking (ICN) is a prominent architecture that realizes content-aware network services through name-based communications. A programmable ICN router implemented on a field-programmable gate array (FPGA) accelerator can achieve predictable performance for advanced service requirements, such as low latency and high throughput. However, owing to the tight resource constraints, implementing ICN functions on an FPGA is a huge challenge, as ICN requires frequent update tables for maintaining long and variable-length content names and access to caching devices capable of storing large-volume content. In this study, we designed and implemented an FPGA router that incorporates an ICN by applying the CCNx v.1.0 protocol specification. The FPGA router implemented a unified status table on a large-capacity DRAM that stored 10 million variable-length content name prefixes. Additionally, we introduced a DRAM bank allocation that is unaffected by slow random accesses and reduces table access latency, and a Row–Bank–Column memory address mapping scheme that reduces the access time to heavily loaded in-network cache storage. We present the evaluation results of the proposed architecture implemented on the NetFPGA-SUME board, revealing 10 Gbps or 2.8 Mpps throughput and ≤ 300 ns jitter.

Abstract:

Programmable switches provide a solution for implementing high-performance Named Data Networking (NDN) routing systems. However, the challenge arises from the packets in the NDN network that carry hierarchical names of varying lengths, coupled with the inherent limitation of programmable switches in parsing variable-length packet content and storing large NDN routing tables. Therefore, deploying NDN routers based on programmable switches in NDN networks remains a formidable task. To address this issue, we implement a parser that supports variable-length names in a programmable switch, and combine the switch with a commodity server cluster to realize a high-speed NDN routing system capable of simultaneous processing names with different lengths at line rate. In this paper, we introduce the design of our routing system in programmable switches and commodity servers, and experimentally verify it can be scaled to Tbps-scale.

Abstract:

The implementation of high-speed Named Networking (NDN) /Content Centric Networking (CCN) forwarders has been a hot research issue over the last decade. Many full-fledged software-based forwarders are implemented toward 100 Gbps throughput, and recently, prototypes of some hardware-based forwarders using a P4 switch are being implemented toward more than 1 Tbps throughput. Many studies press stress on NDN forwarders with uniform request patterns. However, such request patterns only allow the maximum throughput to be measured. On the contrary, this paper addresses the generation of non-uniform request patterns at the rate of Tbps to evaluate the throughput of NDN forwarders.

Abstract:

Logging is crucial for auditing the function and securities of the applications. We designed Mnemosyne, a distributed logger running over Named Data Networking(NDN), to address this need for NDN-based distributed applications. Mnemosyne assures the immutability of logged events by interlocking all event records in a DAG-based blockchain. Mnemosyne provides a high logging throughput and system resiliency against network component failures. We implemented Mnemosyne’s DAG and interface and evaluated our design through an emulated and deployment setting.

Abstract:

The lack of application support is probably the biggest obstacle to ICN/NDN deployment. One approach to tackle this problem is to NDNizing existing applications by translating between application-level protocols and NDN, which can benefit from NDN’s architectural advantages while minimizing development efforts needed. In this paper, we validate the effectiveness of this approach by applying it to Internet livestreaming, and develop PCLive, a livestreaming system with NDN embedded as its distribution network. PCLive makes minimal changes to an Internet livestreaming architecture, achieving the maximum compatibility with existing components including video players, OBS, and video transcoders. By solving a number of design issues such as HLS/NDN protocol translation, data translation, naming and security, PCLive is able to run over NDN network and enjoy its architectural benefits. Since December 2021, PCLive has been running on an NDN testbed consisting of cloud servers from seven cities. It can serve almost four times as many clients as an existing livestreaming system can over IP under the same network conditions; at the same time, the average throughput of the bottleneck link in the NDN testbed is 34.8% lower than that in IP. We also evaluate congestion control and adaptive forwarding with PCLive.

Abstract:

State Vector Sync (SVS) is a Distributed Dataset Synchronization (Sync) protocol designed to support distributed applications running over NDN. The design of SVS has two unique features that set it apart from all the previous Sync protocol designs. First, SVS encodes the raw information of data namespace to be synchronized in its Sync Interest packets. Second, and related, it uses Sync Interests as notifications which do not solicit data replies. To reveal insights of how its unique design features enable SVS to outperform its counterparts, we describe the operation of two types of timers used in SVS and their effectiveness in minimizing protocol overhead while keeping synchronization delay low.

Abstract:

The semantic naming and data-centric security paradigm of Named Data Networking opens up new possibilities for building decentralized applications. In this poster, we briefly identify some factors that lead to today’s centralization and non-interoperability of cloud services. We present the preliminary design of a photo sharing application as a simple example to articulate how NDN can overcome the lack of cloud-independent user identities, illustrating the process and considerations involved in designing decentralized applications that can provide the features offered by existing cloud services.

Abstract:

We present a name-based sidecar-assisted networking approach that enables off-the-shelf TCP/IP and HTTP protocols to effectively handle network disruptions. We compare the performance of our approach with the Name Data Networking by deploying and testing them in a hardware networking testbed and show, on average, in 85% of experiments, the sidecar approach manages to transfer data.

Abstract:

We focus on the security and performance characteristics of the QUIC protocol, exploring its security qualitatively and its performance quantitatively through experimentation. Performance comparisons between QUIC and traditional TCP-based counterparts were made through experiments under various dimensions. These comparisons were based on measurements of running code on real computing and networking resources within Washington University’s Open Network Lab (ONL). The experimental results show that QUIC clients perform as well as or better than TCP clients for single-stream resources with limited bandwidth. However, minimal performance differences exist between QUIC and TCP for multiple streams.

Abstract:

A unique characteristic of LEO satellite networks (satnet) is their dynamic connectivity changes at high frequency, which brings a challenge to Named Data Networking (NDN). As a data-centric network architecture, NDN uses in-network states to steer requested data packets back to consumers. The dynamic connectivity changes of LEO satnets can disrupt the states of NDN’s forwarding plane, raising the question of whether NDN could run on LEO satnets. In this poster, we graphically illustrate how NDN’s stateful forwarding enables multicast data delivery and detects forwarding loops; and how the Content Store (CS) at NDN nodes enables fast recovery of lost data packets and fast delivery for future requests for the same data. We argue that the PIT and CS designed into the NDN forwarding plane can greatly improve a LEO satnet’s data delivery performance. Therefore we must develop effective solutions to maintain NDN’s stateful forwarding plane in the face of frequent dynamic connectivity changes.