In the cloud-native era, developers have at their disposal an unprecedented landscape of services to build scalable distributed systems. The DevOps paradigm emerged as a response to the increasing necessity of better automations, capable of dealing with the complexity of modern cloud systems. For instance, Infrastructure-as-Code tools provide a declarative way to define, track, and automate changes to the infrastructure underlying a cloud application. Assuring the quality of this part of a code base is of utmost importance. However, learning to produce robust deployment specifications is not an easy feat, and for the domain experts it is time-consuming to conduct code-reviews and transfer the appropriate knowledge to novice members of the team. Given the abundance of data generated throughout the DevOps cycle, machine learning (ML) techniques seem a promising way to tackle this problem. In this work, we propose an approach based on Large Language Models to analyze declarative deployment code and automatically provide QA-related recommendations to developers, such that they can benefit of established best practices and design patterns. We developed a prototype of our proposed ML pipeline, and empirically evaluated our approach on a collection of Kubernetes manifests exported from a repository of internal projects at Nokia Bell Labs.

Communication networks need to be both adaptive and scalable. The last few years have seen an explosive growth of software-defined networking (SDN) and network function virtualization (NFV) to address this need. Both technologies help enable networking software to be decoupled from the hardware so that software functionality is no longer constrained by the underlying hardware and can evolve independently. Both SDN and NFV aim to advance a software-based approach to networking, where networking functionality is implemented in software modules and executed on a suitable cloud computing platform. Achieving this goal requires the virtualization paradigm used in these services that play an important role in the transition to software-based networks. Consequently, the corresponding computing platforms accompanying the virtualization technologies need to provide the required agility, robustness, and scalability for the services executed. Serverless computing has recently emerged as a new paradigm in virtualization and has already significantly changed the economics of offloading computations to the cloud. It is considered as a low-latency, resource-efficient, and rapidly deployable alternative to traditional virtualization approaches, such as those based on virtual machines and containers. Serverless computing provides scalability and cost reduction, without requiring any additional configuration overhead on the part of the developer. In this paper, we explore and survey how serverless computing technology can help building adaptive and scalable networks and show the potential pitfalls of doing so.

Virtualization helps to deploy the functionality of expensive and rigid hardware appliances on scalable virtual resources running on commodity servers. However, optimal resource provisioning for non-trivial services is still an open problem. While there have been efforts to answer the questions of when to provision additional resources in a running service, and how many resources are needed, the question of what should be provisioned has not been investigated, in particular, for complex applications or services, which consist of a set of connected components, where each component in turn potentially consists of multiple component instances (e.g., VMs or containers). Each instance of a component can be run in different flavors (i.e., number of cores or amount of memory), while the service constructed by the combination of these component configurations must satisfy the customer Service Level Objective (SLO). In this work, we offer to service providers an answer to the what to deploy question by introducing Rconf, a system that automatically chooses the optimal combination of component instances for non-trivial network services. In particular, we propose an analytical model based on robust queuing theory that is able to accurately model arbitrary components, and develop an algorithm that finds the combination of their instances, such that the overall utilization of the running instances is maximized while meeting SLO requirements.

The virtualization of services with high-availability requirements calls to revisit traditional operation and provisioning processes. Providers are realizing services in software on virtual machines instead of using dedicated appliances to dynamically adjust service capacity to changing demands. Cloud orchestration systems control the number of service instances deployed to make sure each service has enough capacity to meet incoming workloads. However, determining the suitable build-out of a service is challenging as it takes time to install new instances and excessive re-configurations (i.e. scale in/out) can lead to decreased stability. In this paper we present AidOps, a cloud orchestration system that leverages machine learning and domain-specific knowledge to predict the traffic demand, optimizing service performance and cost. AidOps does not require a conservative provisioning of services to cover for the worst-case demand and significantly reduces operational costs while still fulfilling service quality expectations. We have evaluated our framework with real traffic using an enterprise application and a communication service in a private cloud. Our results show up to 4X improvement in service performance indicators compared to existing orchestration systems. AidOps achieves up to 99.985% availability levels while reducing operational costs at least by 20%.

A common practice to increase the reliability of a cloud application is to deploy redundant instances. Unfortunately such redundancy efforts can be undermined if the application's instances share common dependencies. This paper presents ReCloud, a novel system that can efficiently find a reliable deployment plan for cloud applications. ReCloud considers and avoids common dependencies shared across application instances that may lead to correlated failures, and works with applications that even have complex internal structures. ReCloud utilizes various pieces of available dependency information (e.g., hardware, software and/or network dependencies) about the cloud infrastructure to quantitatively assess the reliability of the application's deployment plan with rigorous error bounds. This assessment further enables ReCloud to find a deployment plan that balances between reliability and other criteria such as application performance and resource utilization. We implemented a fully functional system. The experimental results show that, even in a large cloud environment with more than 27K hosts, ReCloud needs only 30 seconds to find a deployment plan that is one order of magnitude more reliable than the common practice.

Network functions virtualization requires automatic deployment and scaling of components. This raises the question of where to place instances of a function, for instance in the OpenStack cloud system. Data plane functions can forward large amounts of traffic. In this case, network-aware placement can avoid an inefficient use of host bandwidth, and a chain of functions can benefit from co-locating instances on a host. However, a practical challenge is that the bandwidth utilization or traffic demand matrix is not always known before the deployment of an instance. A promising remedy is to leverage existing Software Defined Networking (SDN) policies to derive connectivity weights between components. In this paper, we present this novel solution to the online instance placement problem. We have developed an extension of the OpenStack scheduler that uses SDN forwarding policies to rank potential hosts. For a given type of virtual machine, the corresponding forwarding policies can be retrieved from an SDN controller prior to the placement decision. Our prototype identifies potential communication peers and weighs the forwarding rules to prefer hosts that already run communication peers. We present heuristics for such weighing, and we also discuss limitations of the approach. A testbed implementation proofs that even in a simple example our solution can double the service chain throughput.

Onboarding network functions onto current clouds requires labor-intensive configuration of the virtual environment. Developers need to dimension the resources available to each virtual machine such as CPU and memory, define thresholds for scaling dynamically and create configuration files that operators can use to execute the network services. This process is time consuming and dependent on the server architecture. As resources are managed on an individual virtual machine basis, services cannot be orchestrated end to end without significant expertise. In this paper, we argue that much of the manual configuration needed for onboarding services onto a cloud can be automated. Moreover, we can automatically generate abstractions that consider services end-to-end and enable their holistic orchestration. We propose a framework that benchmarks network services during the onboarding process and generates an elastic model which relates workload mixes to resource requirements, identifies component dependencies and automates service operation on heterogeneous stacks. We have evaluated our framework using a real-time communication service that handles multiple classes of workloads. Results show that underprovisioning can be eliminated for regular daily traffic, reducing resource provisioning time by at least 5X for the most stressing traffic surges, while improving key performance indicators by at least 40%.

Approximate computing aims for efficient execution of workflows where an approximate output is sufficient instead of the exact output. The idea behind approximate computing is to compute over a representative sample instead of the entire input dataset. Thus, approximate computing --- based on the chosen sample size --- can make a systematic trade-off between the output accuracy and computation efficiency.

Network functions virtualization (NFV) environments can provide increased elasticity and flexibility for operators, as network-related services can be scaled and moved as needed. Such operations require a seamless transfer of state to provide a service without interruption or performance degradation. In this paper, we propose and analyze a novel approach for efficient and seamless NFV state transfers. Our approach is based on the concept of statelets, which are compact representations of information in incoming packets that change the state of a virtualized network function. We present and describe slim migration (SliM), a system for efficient NFV state transfers using a statelet interface that we have implemented as an add-on to the data plane development kit, a high-performance packet I/O library. We have evaluated SliM in a testbed and present results that show its benefits in terms of lower delays and lower packet-loss rates. Moreover, our analysis and evaluation results show that SliM offers three times as much shared link capacity to the dataplane as previous approaches, while reducing the duration of a state transfer to one-third of their time.

Instance migration and scale-in/out operations in network functions virtualization (NFV) require state transfer mechanisms, which are known to cause service degradation through increased jitter and packet loss. Techniques such as packet duplication for state synchronization mitigate this problem, however, they incur significant additional costs. In this paper, we provide a novel interface to the VNF to announce “statelets” for incoming packets, which comprise only the information in the packet which is required for a VNF's internal state change. Based on this interface, we design and implement SliM, a statelet-based framework for seamless VNF migration. First evaluation results show that SliM operates seamlessly at very high dataplane utilization of physical links, up to 3 times the utilization level at which existing approaches are failing due to insufficient bandwidth for state synchronization.

Cloud computing systems require a placement logic that decides where to allocate resources. In state-of-the-art platforms such as OpenStack, this scheduler takes into account multiple constraints when starting a new instance, including in particular the required computational and memory resources. However, this scheduling mechanism typically neither considers network requirements of Virtual Machines nor the networking resources that are actually available. In this paper we present an extension of the OpenStack scheduler that enables a network-aware placement of instances by taking into account bandwidth constraints to and from nodes. Our solution keeps track of host-local network resource allocation, and it can be combined with bandwidth enforcement mechanisms such as rate limiting. We present a prototype that requires only very few changes in the OpenStack open source software. Testbed measurement results demonstrate the benefit of our solution compared to the OpenStack default approach.

Netflix and Hulu are leading Over-the-Top (OTT) content service providers in the US and Canada. Netflix alone accounts for 29.7% of the peak downstream traffic in the US in 2011. Understanding the system architectures and performance of Netflix and Hulu can shed light on the design of such large-scale video streaming platforms, and help improving the design of future systems. In this paper, we perform extensive measurement study to uncover their architectures and service strategies. Netflix and Hulu bear many similarities. Both Netflix and Hulu video streaming platforms rely heavily on the third-party infrastructures, with Netflix migrating that majority of its functions to the Amazon cloud, while Hulu hosts its services out of Akamai. Both service providers employ the same set of three content distribution networks (CDNs) in delivering the video contents. Using active measurement study, we dissect several key aspects of OTT streaming platforms of Netflix and Hulu, e.g., employed streaming protocols, CDN selection strategy, user experience reporting, etc. We discover that both platforms assign the CDN to a video request without considering the network conditions and optimizing the user-perceived video quality. We further conduct the performance measurement studies of the three CDNs employed by Netflix and Hulu. We show that the available bandwidths on all three CDNs vary significantly over the time and over the geographic locations. We propose a measurement-based adaptive CDN selection strategy and a multiple-CDN-based video delivery strategy that can significantly increase users' average available bandwidth.

The ability to extract topology information from the network is important for many applications and enables more informed resource selection. The challenge for topology exposure is to provide a compact representation that is sufficiently accurate and complies to topology hiding policies. This paper presents a topology abstraction system that can expose large-scale service provider network maps with an adjustable level-of-detail. Our system uses graph sparsification algorithms to reduce the complexity of routing topologies. Our numerical results reveal that the size of maps can be reduced by one order of magnitude or more while the result still enables reasonable traffic optimization inside applications. A proof-of-concept implementation gathers network management system data and exposes abstract maps through the Application-Layer Traffic Optimization (ALTO) protocol.

Swarm-based peer-to-peer streaming (SPS) mechanisms tend to generate a significant amount of costly inter-ISP traffic. Localization of overlay connectivity reduces inter-ISP traffic. However, it can adversely affect the delivered quality. In this paper, we systematically examine the performance of SPS for live video over localized overlays. We identify and discuss the fundamental bottlenecks limiting the stream quality and present OLIVES, an ISP-friendly P2P streaming mechanism for live video. OLIVES maintains a fully localized overlay to reduce the volume of inter-ISP traffic and incorporates a two-tier inter-ISP and intra-ISP scheduling scheme to maximize the delivered quality to individual peers. One important design choice is to perform basic scheduling at a substream level and to use implicit coordination among peers. This allows OLIVES to efficiently detect missing blocks and pull them into the ISP in a timely manner with a minimum in coordination overhead. Furthermore, OLIVES incorporates a shortcutting technique that limits the buffer requirements for each participating peer and effectively reduced the playout latency. Through analysis and extensive simulations, we demonstrate the ability of OLIVES to deliver high-quality streams over localized overlays in various realistic scenarios.

Welcome to the 21stedition of the International Conference on Network Protocols (ICNP) in Göttingen, Germany. We are pleased you can join us in the historic building of the Mathematical Institute at the Georg-August-University, a place that was the home to pioneers such as David Hilbert, Felix Klein, Emmy Noether and Carl-Friedrich Gauss. At this renowned university where the first telegraph was invented by its professors Carl-Friedrich Gauss and Wilhelm Weber and used to communicate with each other using a three kilometer wire between two office buildings 180 years ago, we are looking forward to an inspiring four days of workshops, paper presentations, discussions and social networking among researchers from all over the world with an interest in modern computer networking protocols.

The Software Defined Network (SDN) model depends on abstractions to separate the control plane from the packet forwarding plane. Applications can interact with the control plane to receive a global network view, upon which they can operate. By having access to network topology information, applications can optimize decisions related to service rendezvous, service fulfillment, service placement and service removal. The network is in the best position to provide guidance to a broad class of applications, including peer-to-peer systems, Content Distribution Network (CDN), and datacenter applications. In all these use cases proximity matters as peers need to rendezvous with other peers and users need to rendezvous with the best cloud application or best CDN server. We maintain that a solution for such a rendezvous problem should be an intrinsic component of the emerging SDN model. A specific instance of a protocol that abstracts network topology is the Application Layer Traffic Optimization (ALTO) protocol. ALTO provides applications an abstract view of the network and thus enables applications to leverage a network without exposing the network provider's internal details or policies. We argue that ALTO provides a clean, mature, standards-based and powerful abstraction, which can be used by SDNs today to obtain network information for solving the rendezvous problem.

Content-centric networks promise to address content networking issues in a better way than today's host-based networking architecture. But content-centric networking does not inherently address the issue of services, particularly service scalability and mobility. We present our work on CCNxServ, a system that allows for dynamic service deployment and scalability in a content-centric networking implementation (CCNx) through an intuitive use of the content naming scheme. It thus extends the concept of content-centric networking towards services.

Netflix is the leading provider of on-demand Internet video streaming in the US and Canada, accounting for 29.7% of the peak downstream traffic in US. Understanding the Netflix architecture and its performance can shed light on how to best optimize its design as well as on the design of similar on-demand streaming services. In this paper, we perform a measurement study of Netflix to uncover its architecture and service strategy. We find that Netflix employs a blend of data centers and Content Delivery Networks (CDNs) for content distribution. We also perform active measurements of the three CDNs employed by Netflix to quantify the video delivery bandwidth available to users across the US. Finally, as improvements to Netflix's current CDN assignment strategy, we propose a measurement-based adaptive CDN selection strategy and a multiple-CDN-based video delivery strategy, and demonstrate their potentials in significantly increasing user's average bandwidth.

HTTP adaptive streaming (HAS) is quickly becoming a popular mechanism for delivering on-demand video content over the Internet. The chunked transmission and application-layer adaptation create a very different traffic pattern than traditional progressive video downloads where the entire video is downloaded with a single request.

We consider a system of compute and storage resources geographically distributed over a large number of locations connected via a wide-area network. By distributing the resources, latency to users can be decreased, bandwidth costs reduced and availablility increased. The challenge is to distribute services with varying characteristics among the data centers optimally. Some services are very latency sensitive, others need vast amounts of storage, and yet others are computationally complex but do not require hard deadlines on execution. We propose efficient algorithms for the placement of services to get the maximum benefit from a distributed cloud systems. The algorithms need input on the status of the network, compute resources and data resources, which are matched to application requirements.

Content-centric networks have been proposed as a new network paradigm that is centered around the distribution of content. A key idea of content-centric networks is to address content by name and to enable nodes in the network to respond to content requests. Most proposals for content-centric networks require a "clean slate" approach and a replacement of today's TCP/IP protocol stack, which raises questions about a feasible deployment path. In this paper, we ask the question to which extent the ideas of content-centric networks can be realized on top of today's IP protocol suite. We explore an approach for name-based addressing that extends today's TCP/IP protocols in a fully standard compliant way. We implement our new method in order to demonstrate its feasibility and evaluate the performance of the system using both latency and processing overhead as measures. The obtained results demonstrate that name-based addressing on IP is feasible. We also acknowledge that content-centric networks designed from the ground up go beyond what can be achieved on top of IP.

Although Tor is the most widely used overlay for providing anonymity services, its users often experience very high delays. Because much of Tor usage is for Web applications, which are sensitive to latency, it is critical to reduce delays in Tor. To take an important step in this direction, we seek an in-depth understanding of delays in Tor. By taking snapshots of the entire Tor network within a short time window, we are able to study the delay distribution of the entire router population. We also monitor delays introduced by individual Tor routers over extended periods of time. Our results indicate that apart from delays introduced by routers, overlay network latency also plays a significant role in delays in Tor. We have also observed that at any time, there exist huge differences in the delays introduced by different routers. Our results reveal key performance characteristics of Tor system behavior and provide valuable insights for improving the Tor performance.

The availability of Internet services brings many benefits to developing regions, yet Internet deployment levels in these regions remain staggeringly low. In this work we investigate how existing cellular deployments, which have enjoyed more rapid and wider deployment than client Internet infrastructure, could be used to provide very low cost Internet services in underdeveloped rural areas. We propose a new service model in which traffic is delivered over multihop client-to-client connections that are coordinated by end-to-end control traffic exchanged over cellular infrastructure. To enable this scheme in low client density rural settings, we propose a novel data forwarding mechanism for opportunistic space-time paths. To explore multiple opportunistic paths, but without the high forwarding cost of replicating data on these paths, we use network coding and send only a fraction of the data on each path. Through extensive OPNET simulations we show that globally coordinated opportunistic forwarding enables service acceptable to most applications at only a fraction of cellular infrastructure load. We argue that the reduced load on the cellular infrastructure allows additional users to share services and cost of the network and has the potential to lower the per user price of data services in developing regions.

The session initiation protocol (SIP) is rapidly being adopted as the signaling protocol for establishing, modifying and terminating multimedia sessions. With the increasing use of SIP in large deployments, it is now becoming apparent that the current SIP design does not easily scale up to large network sizes and SIP servers are not well equipped to handle overload conditions. When a SIP server is operating close to or above its capacity limit, message retransmissions by various SIP timers can cause the network to be severely overloaded and result in an extremely low goodput. In this paper, we first provide a detailed analysis of the behavior of SIP servers under overload. We show that SIP servers are often unable to recover from congestion collapse once it has occurred and that overload can spread throughout a network of SIP servers. We then discuss mechanisms and algorithms for controlling overload in these servers. We found that performing overload control locally at a server provides a simple remedy for light cases of overload; however, it is ineffective in handling higher amounts of load. Finally, we investigate distributed overload control mechanisms for SIP and show that they are effective in controlling overload of SIP servers.

Overlay networks are widely used as a way to introduce novel Internet services and to improve the reliability and performance of today's Internet. Traditionally, overlay networks perform routing by mimicking the underlying IP routing protocols, which does not scale well with the system size. In this paper, we propose a fundamentally new approach to overlay routing that is based on network coordinates. Our evaluation shows that this approach is scalable to a large number of overlay nodes with a message complexity of O(N) while providing routing decisions that are close to optimal in terms of error resilience and path delay.

Overlay routing is known to cause undesired instability in a network by operating in a selfish manner. The objectives of overlay routing, such as optimizing end-to-end latency, are often in conflict with the objectives of traffic engineering in the native layer, which is concerned about balancing load. In our work, we build on past research that has investigated the recurring non-cooperative interaction between overlay routing and traffic engineering, and develop strategies that improve the routing performance of a particular layer with incomplete information about the other layer. In our strategies, one layer acts as a leader that predicts the follower's reaction and undertakes countermeasures to prevent future deterioration in performance. Specifically, we propose two classes of strategies - friendly or hostile - for each layer. By simulating under different network characteristics, we show that these preemptive strategies achieve near-optimal performance for the leader and increase the overall stability of the network. Furthermore, we observe that the best performance for a particular layer is achieved only when the goals of the other layer are completely violated, thereby motivating a higher level of selfishness.

Recording systems and media servers for networked audio and video streams have become an important part of today's Internet. In contrast to this, only a few recording and playback solutions currently exist for the data streams of interactive media applications (e.g., shared whiteboards and distributed virtual environments). So far these solutions are application-specific: individual algorithms and implementations are required for each application that is to be recorded. In this paper, we are proposing generic algorithms for the recording and playback of interactive media streams. These algorithms are based on a common model for the class of interactive media. They enable full random access to recordings by initializing the replaying applications with the required state information (e.g., the current slide in a recorded presentation). We have implemented these algorithms in the Interactive Media on Demand (IMoD) system. In order to interpret the semantics of an interactive media stream, the system requires that the Real-Time Application-Level Protocol for Distributed Interactive Media (RTP/I) protocol is used for the framing of the transmitted data. Any application using RTP/I can be recorded directly using the system without any modification. Interactive media streams not using RTP/I can be recorded using the generic recording algorithms. However, they require an adaptation of the system so that it is able to extract a minimal set of information from the application-level protocol of these streams. In addition to the generic recording algorithms, we present the architecture and major design considerations of the system and discuss the experiences gained from recording different interactive media applications.

In this paper, we investigate how consistency can be established for replicated applications changing their state in reaction to user-initiated operations as well as the passing of time. Typical examples of these applications are networked computer games and distributed virtual environments. We give a formal definition of the terms consistency and correctness for this application class. Based on these definitions, it is shown that an important tradeoff relationship exists between the responsiveness of the application and the appearance of short-term inconsistencies. We propose to exploit the knowledge of this tradeoff by voluntarily decreasing the responsiveness of the application in order to eliminate short-term inconsistencies. This concept is called local-lag. Furthermore, a timewarp scheme is presented that complements local-lag by guaranteeing consistency and correctness for replicated continuous applications. The computational complexity of the timewarp algorithm is determined in theory and practice by examining a simple networked computer game. The timewarp scheme is then compared to the well-known dead-reckoning approach. It is shown that the choice between both schemes is application-dependent.

No abstract available.

Distributed interactive media are media that involve communication over a computer network as well as user interactions with the medium itself. Examples of this kind of media are shared whiteboard presentations and networked computer games. One key problem of this media class is that a large amount of common functionality is currently redesigned and redeveloped for each single medium. In order to solve this problem we present a media model and an application level protocol called RTP/I. Derived from the experience gained with audio and video transmission using RTP, RTP/I is defined as a new protocol framework which reuses many aspects of RTP while it is thoroughly adapted to meet the demands of distributed interactive media. By identifying and supporting the common aspects of distributed interactive media RTP/I allows the reuse of key functionality in form of generic services. Furthermore RTP/I makes it possible for applications of different vendors to interact with each other in a standardized way.

In this paper we present a generic late-join service for distributed interactive media, i.e, networked media which involve user interactions. Examples for distributed interactive media are shared whiteboards, networked computer games and distributed virtual environments. The generic late-join service allows a latecomer to join an ongoing session. This requires that the shared state of the medium is transmitted from the old participants of the session to the latecomer in an efficient and scalable way. In order to be generic and useful for a broad range of distributed interactive media, we have implemented the late-join service based on the Real Time Application Level Protocol for Distributed Interactive Media (RTP/I). All applications which employ this protocol can also use the generic late-join service. Furthermore the late-join service can be adapted to the specific needs of a given application by specifying policies for the late-join process. Applications which do use a different application level protocol than RTP/I may still use the concepts presented in this work. However, they will not be able to profit from our RTP/I based implementation.