The introduction of computational resources at the network edge allows application designers to offload computation from clients and/or servers, thereby reducing response latency and backbone bandwidth. More fundamentally, edge-computing moves applications from a client-server model to a client-edge-server model. While this is an attractive paradigm for many use cases, it raises the question of how to design client-edge-server systems so they can tolerate edge failures and client mobility. This is particularly challenging when edge processing is strongly stateful. In this paper we propose a design for meeting this challenge called the Client-Edge-Server for Stateful Network Applications (CESSNA).

This paper is focused not on the Internet architecture - as defined by layering, the narrow waist of IP, and other core design principles - but on the Internet infrastructure, as embodied in the technologies and organizations that provide Internet service. In this paper we discuss both the challenges and the opportunities that make this an auspicious time to revisit how we might best structure the Internet's infrastructure. Currently, the tasks of transit-between-domains and last-mile-delivery are jointly handled by a set of ISPs who interconnect through BGP. In this paper we propose cleanly separating these two tasks. For transit, we propose the creation of a "public option" for the Internet's core backbone. This public option core, which complements rather than replaces the backbones used by large-scale ISPs, would (i) run an open market for backbone bandwidth so it could leverage links offered by third-parties, and (ii) structure its terms-of-service to enforce network neutrality so as to encourage competition and reduce the advantage of large incumbents.

Recent Internet research has been driven by two facts and their contradictory implications: the current Internet architecture is both inherently flawed (so we should explore radically different alternative designs) and deeply entrenched (so we should restrict ourselves to backwards-compatible and therefore incrementally deployable improvements). In this paper, we try to reconcile these two perspectives by proposing a backwards-compatible architectural framework called Trotsky in which one can incrementally deploy radically new designs. We show how this can lead to a permanent revolution in Internet architecture by (i) easing the deployment of new architectures and (ii) allowing multiple coexisting architectures to be used simultaneously by applications. By enabling both architectural evolution and architectural diversity, Trotsky would create a far more extensible Internet whose functionality is not defined by a single narrow waist, but by the union of many coexisting architectures. By being incrementally deployable, Trotsky is not just an interesting but unrealistic clean-slate design, but a step forward that is clearly within our reach.

One of the major concerns about Network Function Virtualization (NFV) is the reduced stability of virtual network functions (VNFs), compared to dedicated hardware appliances. Stateful VNFs make recovery a complex process, where a major concern is how to handle non-determinism such as multi-threaded processing, time dependence, and randomness.

Computing/infrastructure as a service continues to evolve with bare metal, virtual machines, containers and now serverless granular computing service offerings. Granular computing enables developers to decompose their applications into smaller logical units or functions, and run them on small, low cost and short lived computation containers without having to worry about setting up servers - hence the term serverless computing. While serverless environments can be used very cost effectively for large scale parallel processing data analytics applications, it is less clear if network intensive packet processing applications can also benefit from these new computing services as they do not share the same characteristics. This paper examines the architectural constraints as well as current serverless implementations to develop a position on this topic and influence the next generation of computing services. We support our position through measurement and experimentation on Amazon's AWS Lambda service with a few popular network functions.

We present OpenBox — a software-defined framework for network-wide development, deployment, and management of network functions (NFs). OpenBox effectively decouples the control plane of NFs from their data plane, similarly to SDN solutions that only address the network’s forwarding plane. OpenBox consists of three logic components. First, user-defined OpenBox applications provide NF specifications through the OpenBox north-bound API. Second, a logically-centralized OpenBox controller is able to merge logic of multiple NFs, possibly from multiple tenants, and to use a network-wide view to efficiently deploy and scale NFs across the network data plane. Finally, OpenBox instances constitute OpenBox’s data plane and are implemented either purely in software or contain specific hardware accelerators (e.g., a TCAM). In practice, different NFs carry out similar processing steps on the same packet, and our experiments indeed show a significant improvement of the network performance when using OpenBox. Moreover, OpenBox readily supports smart NF placement, NF scaling, and multi-tenancy through its controller.

We present RENE --- a novel encoding scheme for short ranges on Ternary content addressable memory (TCAM), which, unlike previous solutions, does not impose row expansion, and uses bits proportionally to the maximal range length. We provide theoretical analysis to show that our encoding is the closest to the lower bound of number of bits used. In addition, we show several applications of our technique in the field of packet classification, and also, how the same technique could be used to efficiently solve other hard problems such as the nearest-neighbor search problem and its variants. We show that using TCAM, one could solve such problems in much higher rates than previously suggested solutions, and outperform known lower bounds in traditional memory models. We show by experiments that the translation process of RENE on switch hardware induces only a negligible 2.5% latency overhead. Our nearest neighbor implementation on a TCAM device provides search rates that are up to four orders of magnitude higher than previous best prior-art solutions.

This paper starts by demonstrating the vulnerability of Deep Packet Inspection (DPI) mechanisms, which are at the core of security devices, to algorithmic complexity denial of service attacks, thus exposing a weakness in the first line of defense of enterprise networks and clouds. A system and a multi-core architecture to defend from these algorithmic complexity attacks is presented in the second part of the paper. The integration of this system with two different DPI engines is demonstrated and discussed. The vulnerability is exposed by showing how a simple low bandwidth cache-miss attack takes down the Aho-Corasick (AC) pattern matching algorithm that lies at the heart of most DPI engines. As a first step in the mitigation of the attack, we have developed a compressed variant of the AC algorithm that improves the worst case performance (under an attack). Still, under normal traffic its running-time is worse than classical AC implementations. To overcome this problem, we introduce MCA2-Multi-Core Architecture to Mitigate Complexity Attacks, which dynamically combines the classical AC algorithm with our compressed implementation, to provide a robust solution to mitigate this cache-miss attack. We demonstrate the effectiveness of our architecture by examining cache-miss algorithmic complexity attacks against DPI engines and show a goodput boost of up to 73%. Finally, we show that our architecture may be generalized to provide a principal solution to a wide variety of algorithmic complexity attacks.

Similarity search, and specifically the nearest-neighbor search (NN) problem is widely used in many fields of computer science such as machine learning, computer vision and databases. However, in many settings such searches are known to suffer from the notorious curse of dimensionality, where running time grows exponentially with d. This causes severe performance degradation when working in high-dimensional spaces. Approximate techniques such as locality-sensitive hashing [2] improve the performance of the search, but are still computationally intensive.

Deep Packet Inspection (DPI) plays a major role in contemporary networks. Specifically, in datacenters of content providers, the scanned data may be highly repetitive. Most DPI engines are based on identifying signatures in the packet payload. This pattern matching process is expensive both in memory and CPU resources, and thus, often becomes the bottleneck of the entire application. In this paper we show how DPI can be accelerated by leveraging repetitions in the inspected traffic. Our new mechanism makes use of these repetitions to allow the repeated data to be skipped rather than scanned again. The mechanism consists of a slow path, in which frequently repeated strings are identified and stored in a dictionary, along with some succinct information for accelerating the DPI process, and a data path, where the traffic is scanned byte by byte but strings from the dictionary, if encountered, are skipped. Upon skipping, the data path recovers to the state it would have been in had the scanning continued byte by byte. Our solution achieves a significant performance boost, especially when data is from the same content source (e.g., the same website). Our experiments show that for such cases, our solution achieves a throughput gain of 1.25–2.5 times the original throughput, when implemented in software.

Middleboxes play a major role in contemporary networks, as forwarding packets is often not enough to meet operator demands, and other functionalities (such as security, QoS/QoE provisioning, and load balancing) are required. Traffic is usually routed through a sequence of such middleboxes, which either reside across the network or in a single, consolidated location. Although middleboxes provide a vast range of different capabilities, there are components that are shared among many of them.

This paper takes advantage of the emerging multi-core computer architecture to design a general framework for mitigating network-based complexity attacks. In complexity attacks, an attacker carefully crafts "heavy" messages (or packets) such that each heavy message consumes substantially more resources than a normal message. Then, it sends a sufficient number of heavy messages to bring the system to a crawl at best. In our architecture, called MCA2---Multi-Core Architecture for Mitigating Complexity Attacks---cores quickly identify such suspicious messages and divert them to a fraction of the cores that are dedicated to handle all the heavy messages. This keeps the rest of the cores relatively unaffected and free to provide the legitimate traffic the same quality of service as if no attack takes place.

Deep Packet Inspection (DPI) lies at the core of contemporary Network Intrusion Detection/Prevention Systems and Web Application Firewalls. DPI aims to identify various malware (including spam and viruses) by inspecting both the header and the payload of each packet and comparing it to a known set of patterns. DPI is often performed on the critical path of the packet processing, thus the overall performance of the security tools is dominated by the speed of DPI. The seminal algorithm of Aho-Corasick (AC) is the de facto standard for pattern matching in network intrusion detection systems (NIDS). Basically, the AC algorithm constructs a Deterministic Finite Automaton (DFA) for detecting all occurrences of a given set of patterns by processing the input in a single pass. The input is inspected symbol by symbol (usually each symbol is a byte), such that each symbol results in a state transition. Thus, in principle, the AC algorithm has deterministic performance, which does not depend on specific input and therefore is not vulnerable to algorithmic complexity attacks, making it very attractive to NIDS systems. In this paper we show that, when implementing the AC algorithm in software, this property does not hold, due to the fact that contemporary pattern sets induce very large DFAs that cannot be stored entirely in cache. We then propose a novel technique to compress the representation of the Aho-Corasick automaton, so it can fit in modern cache. We compare both the performance and the memory footprint of our technique to previously-proposed implementation, under various settings and pattern sets. Our results reveal the space-time tradeoffs of DPI. Specifically, we show that our compression technique reduces the memory footprint of the best prior-art algorithm by approximately 60%, while achieving comparable throughput.