Today, engineering interdomain traffic in large transit ASs is a difficult task due the opacity o... more Today, engineering interdomain traffic in large transit ASs is a difficult task due the opacity of BGP and the interactions between the BGP decision process and IGP routing. In this paper we propose Tweak-it, a tool that, based on the steady-state view of BGP routing inside the AS and the traffic demands of the AS, computes the BGP updates to be sent to the ingress routers of a transit AS to traffic engineer its interdomain traffic over time. Tweak-it is based on two components: 1) a scalable BGP simulator (CBGP) that computes the steady-state behavior of BGP routing and 2) a multiple-objectives evolutionary heuristic that can deal with multiple conflicting objectives as they can occur in real networks. Tweak-it takes the intradomain configuration (IGP weights and topology), BGP messages received from peers, BGP routing policies, and traffic demands. By keeping the state of the routing inside the AS up-to-date and based on the traffic demands, the heuristics computes how to engineer the traffic of the AS while trying to minimize the number of BGP tweakings required.
Over the past decade, IXPs have been playing a key role in enabling interdomain connectivity. The... more Over the past decade, IXPs have been playing a key role in enabling interdomain connectivity. Their traffic volumes have grown dramatically and their physical presence has spread throughout the world. While the relevance of IXPs is undeniable, their long-term contribution to the shaping of the current Internet is not fully understood yet. In this paper, we look into the impact on Internet routes of the intense IXP growth over the last decade. We observe that while in general IXPs only have a small effect in path shortening, very large networks do enjoy a clear IXP-enabled path reduction. We also observe a diversion of the routes, away from the central Tier-1 ASes supported by IXPs. Interestingly, we also find that whereas IXP membership has grown, large and central ASes have steadily moved away from public IXP peerings, whereas smaller ones have embraced them. Despite all this changes, we find though that a clear hierarchy remains, with a small group of highly central networks
Proceedings of the 11th ACM Conference on Emerging Networking Experiments and Technologies
TCP is an important factor affecting user-perceived performance of Internet applications. Diagnos... more TCP is an important factor affecting user-perceived performance of Internet applications. Diagnosing the causes behind TCP performance issues in the wild is essential for better understanding the current shortcomings in TCP. This paper presents a TCP flow performance analysis framework that classifies causes of TCP stalls. The framework forms the basis of a tool that is publicly available to the research community. We use our tool to analyze packet-level traces of three services (cloud storage, software download and web search) deployed by a popular Chinese service provider. We find that as many as 20% of the flows are stalled for half of their lifetime. Network-related causes, especially timeout retransmission, dominate the stalls. A breakdown of the causes for timeout retransmission stalls reveals that double retransmission and tail retransmission are among the top contributors. The importance of these causes depends however on the specific service. We also propose S-RTO, a mechanism that mitigates timeout retransmission stalls. S-RTO has been deployed on production front-end servers and results show that it is effective at improving TCP performance, especially for short flows.
Hypergiants, such as Google or Netflix, are important organisations in the Internet ecosystem, du... more Hypergiants, such as Google or Netflix, are important organisations in the Internet ecosystem, due to their sheer impact in terms of traffic volume exchanged. However, the research community still lacks a sufficiently crisp definition for them, beyond naming specific instances of them. In this paper we analyse PeeringDB data and identify features that differentiate hypergiants from the other organisations. To this end, we first characterise the organisations present in PeeringDB, allowing us to identify discriminating properties of the these organisations. We then use these properties to separate the data in two clusters, differentiating hypergiants from other organisations. We conclude this paper by investigating how hypergiants and other organisations exploit the IXP ecosystem to reach the global IPv4 space.
Proceedings of the 2018 ACM SIGSIM Conference on Principles of Advanced Discrete Simulation
Assessing the impact of changes in a production network (e.g., new routing protocols or topologie... more Assessing the impact of changes in a production network (e.g., new routing protocols or topologies) requires simulation or emulation tools capable of providing results as close as possible to those from a real-world experiment. Large traffic loads and complex controldata plane interactions constitute significant challenges to these tools. To meet these challenges we propose a model for the fast and convenient evaluation of SDN as well as legacy networks. Our approach emulates the network's control plane and simulates the data plane, to achieve high fidelity necessary for control plane behavior, while being capable of handling large traffic loads. We design and implement a proof of concept from the proposed model. The initial results of the prototype, compared to a state-of-the-art solution, shows it can increase the speed of network experiments by nearly 95% in the largest tested network scenario.
Over the past decade, Internet eXchange Points (IXPs) have been playing a key role in enabling in... more Over the past decade, Internet eXchange Points (IXPs) have been playing a key role in enabling interdomain connectivity. Their traffic volumes have quickly become similar to those of Tier-1 ASes and their physical presence has spread across the world. While the relevance of IXPs is undeniable, their contribution to the shaping of the current Internet is not fully understood yet, especially on the so-called "flattening" phenomenon. In this paper, we disentangle "flattening" from "path-shortening": we show that the impact of IXPs on path length is limited. Surprisingly, Internet path-lengths have globally barely decreased over the last decade, regardless of whether they cross IXPs or not. Instead of a "path-shortening", we observe a diversion of the routes away from the central Tier-1 ASes, supported by IXPs. This diversion has however not fundamentally changed the existence of a hierarchy, but rather replaced the central Tier-1s with a new set ...
The importance of IXPs to interconnect di↵erent networks and exchange tra c locally has been well... more The importance of IXPs to interconnect di↵erent networks and exchange tra c locally has been well studied over the last few years. However, far less is known about the role IXPs play as a platform to enable large-scale content delivery and to reach a world-wide customer base. In this paper, we study the infrastructure deployment of a content hypergiant, Netflix, and show that the combined worldwide IXP substrate is the major corner stone of its Content Delivery Network. This highlights the additional role that IXPs play in the Internet ecosystem, not just in terms of interconnection, but also allowing players such as Netflix to deliver significant amounts of tra c.
Social Web 2.0 features have become a vital component in a variety of multimedia systems, e.g., L... more Social Web 2.0 features have become a vital component in a variety of multimedia systems, e.g., Last.fm, Flickr and Spotify. Interestingly, adult video websites are also starting to adopt these Web 2.0 principles, giving rise to the term ``Porn 2.0''. This paper examines a large Porn 2.0 social network, through data covering 563k users. We explore a number of unusual behavioural aspects that set this apart from more traditional multimedia social networks, including differences in browsing activity, social communications and relationship creation. We also analyse the nature and behaviour of content sharing, highlighting the role it plays in the Porn 2.0 community, as well as the preferences that users have when deciding what to consume. We particularly explore the impact that gender and sexuality have on these issues, showing their vital importance for aspects such as profile popularity.
2019 Network Traffic Measurement and Analysis Conference (TMA)
Congestion control algorithms are crucial in achieving high utilization while preventing overload... more Congestion control algorithms are crucial in achieving high utilization while preventing overloading the network. Over the years, many different congestion control algorithms have been developed, each trying to improve over others in specific situations. However, their interactions and coexistence has, to date, not been thoroughly evaluated, which is the focus of this paper. Through head-to-head comparisons of loss-based, delay-based and hybrid types of congestion control algorithms, we reveal that fairness in resources claimed is often not achieved, especially when flows sharing a link have different round-trip times or belong to different groups.
2017 Network Traffic Measurement and Analysis Conference (TMA), 2017
Middleboxes are widely used in today's Internet, especially for security and performance. Mid... more Middleboxes are widely used in today's Internet, especially for security and performance. Middleboxes classify, filter and shape traffic, therefore interfering with application performance and performing new network functions for end hosts. Recent studies have uncovered and studied middleboxes in different types of networks. In this paper, we exploit a large-scale proxy infrastructure, provided by Laminati, to detect HTTP-interacting middleboxes across the Internet. Our methodology relies on a client and server side, to be able to observe both directions of the middlebox interaction. Our results provide evidence for middleboxes deployed across more than 1000 ASes. We observe various middlebox interference in both directions of traffic flows, and across a wide range networks, including mobile operators and data center networks.
1Jackson School of Geosciences, University of Texas at Austin, Austin, USA 2Indian Institute of T... more 1Jackson School of Geosciences, University of Texas at Austin, Austin, USA 2Indian Institute of Tropical Meteorology, Ministry of Earth Sciences, India 3Krea University, Sri City, India 4Department of Systemics, School of Computer Science, University of Petroleum and Energy Studies, Dehradun, India 5School of Electronic Engineering and Computer Science, Queen Mary University of London, Mile End Rd, Bethnal Green, London E1 4NS, UK
Congestion control algorithms are crucial in achieving high utilization while preventing overload... more Congestion control algorithms are crucial in achieving high utilization while preventing overloading the network. Over the years, many different congestion control algorithms have been developed, each trying to improve in specific situations. However, their interactions and co-existence has, to date, not been thoroughly evaluated, which is the focus of this paper. Through head-to-head comparisons of representatives from loss-based, delay-based and hybrid types of congestion control algorithms, we reveal that fairness in resources claimed is often not attained, especially when flows sharing a link have different RTTs.
Under increasing scrutiny, many web companies now offer bespoke mechanisms allowing any third par... more Under increasing scrutiny, many web companies now offer bespoke mechanisms allowing any third party to file complaints (e.g., requesting the de-listing of a URL from a search engine). While this self-regulation might be a valuable web governance tool, it places huge responsibility within the hands of these organisations that demands close examination. We present the first large-scale study of web complaints (over 1 billion URLs). We find a range of complainants, largely focused on copyright enforcement. Whereas the majority of organisations are occasional users of the complaint system, we find a number of bulk senders specialised in targeting specific types of domain. We identify a series of trends and patterns amongst both the domains and complainants. By inspecting the availability of the domains, we also observe that a sizeable portion go offline shortly after complaints are generated. This paper sheds critical light on how complaints are issued, who they pertain to and which domains go offline after complaints are issued.
Proceedings of the 2018 Conference of the ACM Special Interest Group on Data Communication
When network is undergoing problems such as congestion, scan attack, DDoS attack, etc., measureme... more When network is undergoing problems such as congestion, scan attack, DDoS attack, etc., measurements are much more important than usual. In this case, traffic characteristics including available bandwidth, packet rate, and flow size distribution vary drastically, significantly degrading the performance of measurements. To address this issue, we propose the Elastic sketch. It is adaptive to currently traffic characteristics. Besides, it is generic to measurement tasks and platforms. We implement the Elastic sketch on six platforms: P4, FPGA, GPU, CPU, multi-core CPU, and OVS, to process six typical measurement tasks. Experimental results and theoretical analysis show that the Elastic sketch can adapt well to traffic characteristics. Compared to the state-of-the-art, the Elastic sketch achieves 44.6 ∼ 45.2 times faster speed and 2.0 ∼ 273.7 smaller error rate.
Click-based network function (NF) has significant advantages for pipeline development, including ... more Click-based network function (NF) has significant advantages for pipeline development, including modularity, extensibility, and programmability. Despite these features, its internal architecture has unfortunately not kept up with some specific problems of the software upgrade. To motivate our article, we analyzed a series of use cases to identify the limitations of native Click. These limitations include the inefficiencies in modifying modules, integrating modules, and recovering states. To bridge the gap, we present three novel enhancements in our Click upgrade (Click-UP) system: 1) modular state abstraction refines each type of stateful operations as an atom operation and decouples it from the pipeline, letting separately managing logics for stateless operations and stateful operations become practical; 2) essential module integration manages dependencies between modules, avoiding shipping unnecessary modules with neutral functionalities to the target NF; and 3) local state migration migrates needed states seamlessly from the old NF to the target NF at local memory. Our evaluation demonstrates that Click-UP reduces the context code required for module modification by 12-81%, cutting down the NF integration time by 78-96% and the service disruption time by 76-93%, as compared to the software upgrade performance represented by native Click.
When network is undergoing problems such as congestion, scan attack, DDoS attack, etc, measuremen... more When network is undergoing problems such as congestion, scan attack, DDoS attack, etc, measurements are much more important than usual. In this case, traffic characteristics including available bandwidth, packet rate, and flow size distribution vary drastically, significantly degrading the performance of measurements. To address this issue, we propose the Elastic sketch. It is adaptive to currently traffic characteristics. Besides, it is generic to measurement tasks and platforms. We implement the Elastic sketch on six platforms: P4, FPGA, GPU, CPU, multi-core CPU, and OVS, to process six typical measurement tasks. Experimental results and theoretical analysis show that the Elastic sketch can adapt well to traffic characteristics. Compared to the state-of-the-art, the Elastic sketch achieves 44.6 ∼ 45.2 times faster speed and 2.0 ∼ 273.7 smaller error rate.
Hypergiants, such as Google or Netflix, are important organisations in the Internet ecosystem, du... more Hypergiants, such as Google or Netflix, are important organisations in the Internet ecosystem, due to their sheer impact in terms of traffic volume exchanged. However, beyond naming specific instances, the research community still lacks a sufficiently crisp understanding of them. In this paper we analyse PeeringDB data and identify features that differentiate hypergiants from the other organisations. To this end, we first characterise the organisations present in PeeringDB, allowing us to identify discriminating properties of these organisations. We then use these properties to separate the data in two clusters, differentiating hypergiants from other organisations. We conclude this paper by investigating how hypergiants and other organisations exploit the IXP ecosystem to reach the global IPv4 space.
Approximate stream processing algorithms, such as Count-Min sketch, Space-Saving, support numerou... more Approximate stream processing algorithms, such as Count-Min sketch, Space-Saving, support numerous applications across multiple areas such as databases, storage systems, and networking. However, the unbalanced distribution in real data streams are challenging to existing algorithms. To enhance these algorithms, we propose a meta-framework, called Cold Filter, that enables faster and more accurate stream processing. Different from existing filters that mainly focus on hot (frequent) items, our filter captures cold (infrequent) items in the first stage, and hot items in the second stage. Existing filters also require twodirection communication-with frequent exchanges between the two stages; our filter on the other hand is one-direction-each item enters one stage at most once. Our filter can accurately estimate both cold and hot items, providing a level of genericity that makes it applicable to many stream processing tasks. To illustrate the benefits of our filter, we deploy it on four typical stream processing tasks. Experimental results show speed improvements of up to 4.7 times, and accuracy improvements of up to 51 times.
The importance of IXPs to interconnect different networks and exchange traffic locally has been w... more The importance of IXPs to interconnect different networks and exchange traffic locally has been well studied over the last few years. However, far less is known about the role IXPs play as a platform to enable large-scale content delivery and to reach a world-wide customer base. In this paper, we study the infrastructure deployment of a content hypergiant, Netflix, and show that the combined worldwide IXP substrate is the major corner stone of its Content Delivery Network. This highlights the additional role that IXPs play in the Internet ecosystem, not just in terms of interconnection, but also allowing players such as Netflix to deliver significant amounts of traffic.
Uploads
Papers by Steve Uhlig