Inhalt des Dokuments
- 2014/12/12 – Master student introductory talk: Stefan Gerganov, "Towards Improved Internet Infrastructure Geolocation"
- 2014/12/05 – Presentation: Marcin Bienkowski, "Online Control Message Aggregation in Chain Networks"
- 2014/12/02 – Master student talk: Carlos Mayer, "User driven traffic prioritization in WiFi networks through participatory networking, following an SDN and NFV approach."
- 2014/11/18 – Talk: Thomas Krenc and Tobias Jacobowitz, "Season"
- 2014/11/18 – Masterstudent talk: Patrick Kutter, "Improving Video STreaming QoE Through Multi Access Policies"
- 2014/10/14 – Talk: Thomas Krenc, "Analysis of public BGP routing tables"
- 2014/07/29 – Introductory talk: Thorben Krüger
- 2014/07/14 – Talk: Srivatsan Ravi, "Inherent Limitations of Hybrid Transactional Memory"
- 2014/06/24 – PGT talk: Felix Poloczek, "Stochastic Network Calculus with Martingales"
- 2014/06/12 – PGT Talk Dan Levin, "Panopticon: Reaping the Beneﬁts of Incremental SDN Deployment in Enterprise Networks"
- 2014/06/10– PGT: Julius Zander - Schulz, "Programmatic Orchestration of WiFi Networks"
Friday, 12. December 2014
|Speaker:||Stefan Gerganov |
|Type:||Master student introductory talk|
|Time:||12 December 2014|
Geolocation, the mapping of IP addresses to geographical locations, continues to receive significant attention both from the academic community as well as from commercial organizations. Most available geolocation databases aim at providing precise locations of IP addresses belonging to end-users but are unable to geolocate IP addresses assigned to infrastructure such as routers and servers. This thesis aims at developing an approach to geolocate IP addresses with a particular emphasis on locating Internet infrastructure. The approach attempts to increase accuracy of RTT measurements and primarily builds on speed-of-light latency constraints using the CAIDA Ark infrastructure; a globally distributed set of probing devices which continuously traceroute the entire IPv4 address space.
Friday, 05. December 2014
|Speaker:||Marcin Bienkowski (Wroclaw University)|
|Time:||5 December 2014|
In the Control Message Aggregation problem, control packets are generated over time at the nodes of a tree T and need to be transmitted to the root of T. To optimize the overall cost, these transmissions can be delayed and different packets can be aggregated, that is a single transmission can include all packets from a subtree rooted at the root of T. The objective is to compute a schedule of transmissions with minimum cost, where the cost of a schedule is defined as the sum of all the transmission costs and delay costs of all packets. The problem is known to be NP-hard, even for trees of depth 2. In the online scenario, it is an open problem whether a constant-competitive algorithm exists. I will address the special case of the problem when T is a chain network. For the online case, I will present a 5-competitive algorithm and give a lower bound of 3.618, improving the previous bounds of 8 and 2, respectively.
Tuesday, 02. December 2014
|Speaker:||Carlos Mayer (TU Berlin master student)|
|Type:||Master student talk|
|Time:||02 December 2014|
"The quickly growing demand for Wifi networks and the numerous application-specific requirements stand in stark contrast to today's inflexible and application-oblivious network management and operation in home networks. In this thesis, we develop and evaluate a middlebox interface following an SDN and NVF approach to bring these benefits into the user’s premises. We interface the system with existing SDN interfaces such as OpenFlow and Odin in order to effectively differentiate between and prioritize different applications in wired and WiFi networks. Furthermore, we develop a participatory like interface allowing users to manage application preferences through an SDN application."
Tuesday, 18. November 2014
|Speaker:||Thomas Krenc , Tobias Jacobowitz (TU Berlin)|
|Time:||18 November 2014|
Currently deployed client-server architectures for Massively Multiplayer Online Games (MMOG) serve millions of concurrent clients. Scaling MMOGs to this magnitude is challenging. Several techniques have been proposed in order to distribute client’s load on multiple servers, e.g. spatial decomposition of the virtual world, or ’sharding’, to name some. Most of existing solutions are static in the sense that resources are allocated statically to the system, and cannot be changed during the runtime; more severely, often even the matching of players to servers cannot be changed. As a consequence of this inflexibility, a system either needs to be conservatively overprovisioned (potentially wasting resources and energy), or risk an unpredictable and poor performance under dynamic demand. This work proposes Season, a dynamic and distributed approach based on a scalable overlay network for MMOG servers. Season allows to elastically scale up and down compute resources depending on the demand of clients. Moreover, it introduces an adaptive load-balancing by flexibilizing the assignment of clients to servers preserving the clients area of interest. We evaluate the performance of Season by using extensive simulations, and study the benefits and limitations of adaptive MMOG server overlays.
Tuesday, 18. November 2014
|Speaker:||Patrick Kutter (TU Berlin master student)|
|Time:||18 November 2014|
The talk will cover my Master project and the introduction to my Master thesis as well. The project's goal was to modify a video streaming client using Dynamic Adaptive Streaming over HTTP(DASH) in a way that it uses the Multi ACCess Framework(MUACC), which was developed at FG-INET. Also several improvement to the framework itself were made in the course of this project. The framework is based on a modified Socket API that uses Socket Intents as additional information from the application to create and manage the sockets on different network interfaces. The purpose of the thesis is to further develop the framework and the client, focusing on creating useful policies in combination with the use of Socket Intents in the streaming application. Different kind of Intents and policies will be evaluated according to effectivity and overall video quality within the streaming player.
Tuesday, 14. October 2014
|Time:||14 October 2014|
In this talk I propose a prefix classification that is based on previous work, and is applied on BGP dumps from the past 8 years.
A little background:
Among other things, prefixes can be delegated from one AS to another. A prefix is delegated if a covering/less specific prefix exists which is announced by a different AS. This can be observed in BGP dumps and is usually explained by the use of multi-homing or load balancing. However, in this work some prefixes are classified as indirectly delegated, in which case the two announcing ASes appear to not be connected by an AS-link. While this can be related to e.g. missing links or prefix aggregation, some cases seem to have a different purpose. As a result of my analysis, I want to introduce some findings and characteristics of BGP routing tables that, for the best of my knowledge, are not documented.
Tuesday, 29. July 2014
|Time:||29 July 2014|
For my introductory talk on Tuesday, I will give an overview of research and projects I did during my academic career. First, I will give a review my Bachelor's Thesis about "Unknown word categorization" in the context of Computational Linguistics, explain the approaches and their weaknesses and what I would have done differently in retrospect. Then, I am going to talk about my MSc graduation project about benchmarking Daniel Bernstein's proposed UDP-based secure transport layer protocol "CurveCP". I will concentrate on my experimental design and the tools I developed, again pointing out any flaws that became apparent since. Finally there would be an opportunity to discuss additional topics such as private projects etc.
Tuesday, 01. July 2014
|Time:||1 July 2014|
Several Hybrid Transactional Memory (HyTM) schemes have recently been proposed to complement the fast, but best-effort, nature of Hardware Transactional Memory (HTM) with a slow, reliable software backup. However, the fundamental limitations of building a HyTM with nontrivial concurrency between hardware and software transactions are still not well understood.
In this paper, we propose a general model for HyTM implementations, which captures the ability of hardware transactions to buffer memory accesses, and allows us to formally quantify and analyze the amount of overhead (instrumentation) of a HyTM scheme. We prove the following: (1) it is impossible to build a strictly serializable HyTM implementation that has both uninstrumented reads and writes, even for weak progress guarantees, and (2) under reasonable assumptions, in any opaque progressive HyTM, a hardware transaction must incur instrumentation costs linear in the size of its data set. We further provide two upper bound implementations whose instrumentation costs are optimal with respect to their progress guarantees. In sum, this paper captures for the first time an inherent trade-off between the degree of concurrency a HyTM provides between hardware and software transactions, and the amount of instrumentation overhead the implementation must incur.
Tuesday, 24. June 2014
|Time:||24 June 2014|
The Stochastic Network Calculus (SNC) was developed as an alternative to the classical Queueing Theory. Its main advantage is that by sacrificing some accuracy, i.e. by providing the performance metrics only in terms of (upper) bounds, it is able to handle a much broader class of queueing systems. However, as those bounds are typically derived by the use of very simplistic mathematical tools they can be off by several orders of magnitude making the relevance of SNC seem doubtful. In this talk we propose a novel method to overcome these limitations. It consists of a characterization of queueing systems by certain stochastic processes called martingales. We will see that those martingales allow for a simple handling of typical queueing operations: Flows' multiplexing translates into multiplication and scheduling translates into time-shifing of the corresponding martingales. Finally, by appying a certain inequality which holds specifically for those processes, per-flow delay bounds are d erived for various scheduling policies which - unlike state-of-the-art results - are numerically tight.
Thursday, 12. June 2014
|Time:||12 June 2014|
"The operational challenges posed in enterprise networks present an appealing opportunity for automated orchestration by way of Software-Defined Networking (SDN). The primary challenge to SDN adoption in the enterprise is the deployment problem: How to deploy and operate a network consisting of both legacy and SDN switches, while benefiting from simplified management and enhanced flexibility of SDN. This paper presents the design and implementation of Panopticon, an architecture for operating networks that combine legacy and SDN switches. Panopticon exposes an abstraction of a logical SDN in a partially upgraded legacy network, where SDN benefits can extend over the entire network. We demonstrate the feasibility and evaluate the efficiency of our approach through both testbed experiments with hardware switches and through simulation on real enterprise campus network topologies entailing over 1500 devices. Our results suggest that when as few as 10% of distribution switches support SDN, most of an enterprise network can be operated as a single SDN while meeting key resource constraints. The dry run is presented in a 17 minute conference format with 3 minutes of Q/A."
Tuesday, 10. June 2014
|Speaker:||Julius Zander - Schulz|
|Type:||Project Group Meeting (PGT)|
|Time:||10 June 2014|
"With wireless technologies becoming prevalent at the last hop, today’s network operators need to manage WiFi access networks in unison with their wired counterparts. However, the non-uniformity of feature sets in existing solutions and the lack of programmability makes this a challenging task. This talk presents Odin, an SDN-based solution to bridge this gap. With Odin, we make the following contributions: (i) Light Virtual Access Points (LVAPs), a novel programming abstraction for addressing the IEEE 802.11 protocol stack complexity, (ii) a design and implementation for a software-defined WiFi network architecture based on LVAPs, and (iii) a prototype implementation on top of commodity access point hardware without modifications to the IEEE 802.11 client, making it practical for today’s deployments. To highlight the effectiveness of the approach we demonstrate six WiFi network services on top of Odin including load-balancing, mobility management, jammer detection, automatic channel - selection, energy management, and guest policy enforcement. To further foster the development of our framework, the Odin prototype is made publicly available."