direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

Offene Themen

Extending Big Data Analytics to the Wide-Area

Montag, 02. Mai 2016

Type: large (Master-) Project (PJ), Bachelor-Thesis (BA)

Yesterday's data volume could be analyzed on single specialized machines. Today's data volume can only be analyzed by clusters of machines deployed in centralized data centers. What about tomorrow's data volume?

With today's Big Data Analytics (BDA) we can analyze data produced at run-time in the form of data streams. In this scenario it can happen that the volume of the data stream becomes so large that the BDA system cannot process it quick enough. 

Currently there are two solutions to the problem: (1) Further computing nodes are added to the BDA till either the data can be processed again, or the available resources are exhausted. (2) The part of the BDA pipeline that is lacking behind (usually the node with the most complex operation) excerts Backpressure on the pipeline. The backpressure forces all earlier nodes in the pipeline to accept lower volumes of the data and usually leads to dropping data at the input or source point of the BDA.

In this thesis we want to develop point 2 further. We want to extend the BDA framework Apache Flink to excert Backpressure on the original sources of the data. We assume this sources to be deployed in the Wide Area (e.g. a production facilities or logistic centers). We base the work on the Jetstream project, where data was aggregated according to the varying performance of the network connection between data source and data center.

Requirements:

  • Good Coding skills
  • Able to at least read Java, Python and C++ code
  • Knowledge of the Network Stack

Ressources:

  • The paper: Aggregation and Degradation in JetStream: Streaming analytics in the wide area
  • Jetstream Code: https://github.com/princeton-sns/jetstream
  • Apache Flink: flink.apache.org

Contact: Niklas[at]inet.tu-berlin.de

Bringing OpenFlow Meter to OpenVSwitch

Montag, 02. Mai 2016

Type: large (Master-) Project (PJ), Bachelor-Thesis (BA)

The Open vSwitch is the most famous virtual switch for using SDN and OpenFlow. Open vSwitch can operate both as a soft switch running within the hypervisor, and as the control stack for switching silicon. The kernel datapath is distributed with Linux, and packages are available for Ubuntu, Debian, and Fedora. The software is open source and the latest code can be found on Github.

Although it has full OpenFlow protocol support up to version 1.5, the meter functionality is not implemented on the switch. (See here) The protocol implementation of meters can be found in the ofproto.c file.

However, the ofsoftswich13 by the Brazilian CPqD group has this feature implemented. Their code can also be found on Github. Goal of the project is to implement the meter functionality in the userspace of Open vSwitch by using the code from ofsoftswitch13. The corresponding code for the meter functionality can be found in the udatapath folder on Github.

Requirements:

  • Good knowledge of C
  • Knowledge of the Network Stack
  • Rough understanding of OpenFlow
  • Experience with Mininet would be helpful

Contact: Niklas[at]inet.tu-berlin

Route Server Control Plane Monitoring and Optimization

Dienstag, 27. Oktober 2015

Typ: kleines (Bachelor-) Projekt (PJ), Bachelor-Arbeit (BA), Praktikum (Internship)

Background

IXPs are shared peering platforms that operate a switching fabric to interconnect its customers’ networks. Among the customer that exchange traffic are typically a wide range of network types, e.g., large ISPs, regional providers, hosters, and content providers. Many IXPs offer so-called route servers which greatly simplify the BGP session management for their connected customers. Therefore, route servers collect routing information in a centralized manner and redistribute them to connected customers.

A customer connects to the route server via a single BGP session to set up BGP peering with all other IXP customers that peer with the route server. Clearly, this has a number of benefits such as lowering the maintenance overhead. However, for a variety of reasons IXPs implement filters so that not all incoming routing information is redistributed to all connected customers. In addition, some customers restrict their routing information with respect to traffic engineering purposes, e.g., applying specific filters. 

This leads to IXP customers being not aware of the reason for the routing decisions of the route server, i.e., best path selection, hidden paths, and filtered announcements. In fact, it remains a challenging task to resolve such unwanted restrictions without automated monitoring systems that combine information from different sources.

Task

The goal of this thesis is to shed light on the usage of different filter types influencing route server decisions. This open question can be broken down in specific task and possibly extended according to the applicant’s experience in the relevant fields:

1. Information derived from the DE-CIX route server shall be stored in a database and be query-able by a simple interface (i.e., CLI or web site). Technologies such as Python, Java, and Java-Script combined with a database shall be used.

2. Based on the data collected in the previous step, an analysis of the different filter types shall be executed. It shall answer questions such as: Do certain customer groups use different filter types? Does the use of a certain filter type change over time? Are different filter types typically combined?

Ansprechpartner: Christoph Dietzel (cdietzel@inet.tu-berlin.de), +49 69 1730902

Algorithms for Virtual Network Embeddings

Typ:  großes (Master-) Projekt (PJ), Master-Arbeit (MA), Diplomarbeit (DA)

In this theoretical thesis, we want to study algorithms for the embedding of virtual networks (that is:  guest graphs) into physical networks (that is: host graphs). In contrast to classic graph embedding problems, the virtual network embedding problems introduces additional flexibilities on where to place nodes.

This thesis is theoretic in nature, and we are looking for a student with a good background in theoretical computer science and algorithms.

However, we leave open how the virtual network embedding problem is tackled. Depending on your skills, one of the following approaches could be used:

-       Mixed Integer Programming

-       SAT solver approaches

-       Approximation Offline Algorithms and Competitive Online Algorithms

-       Evaluation of different heuristics by simulation

Interested? Contact us for more details, so we can find the project that best matches your skills/interests. 


Ansprechpartner: Stefan Schmid

A Better Scheduler for Hadoop

Typ: Master-Arbeit (MA), Diplomarbeit (DA)

Hadoop is an open-source Map/Reduce implementation. In this project, we would like to improve the performance of Hadoop using the scheduler we have developed in our group.

We are looking for a student who is not afraid to get her or his hands dirty and experiment with different algorithms in our local compute cluster.

Interested? Contact us for more details.


Ansprechpartner: Stefan Schmid

Router Table Compression

Typ: Master-Arbeit (MA), Diplomarbeit (DA)

Router memory is very expensive today. This raises the question of how to compress forwarding tables efficiently. In this project, you will develop and formally analyze algorithms (optimal algorithms, online algorithms, or approximation algorithms) to compress router tables, e.g., in an OpenFlow context. Depending on your interests, we may also develop a small simulation framework to evaluate your algorithms with `realistic' data.

Interested? Contact us for more details.


Ansprechpartner: Stefan Schmid

Robust Control Plane for Software-Defined Networks

Typ:  großes (Master-) Projekt (PJ), Master-Arbeit (MA), Diplomarbeit (DA)


Software-Defined Networking (SDN) is a new networking paradigm which allows to manage a set of switches and their behavior from a logically centralized perspective: a software controller.

In this project, we want to investigate how to implement a robust SDN control plane which tolerates controller failures and which supports concurrent network updates: Policy changes which have been issued concurrently by different network operators should be composed and installed consistently.

For this project, we are looking for a student with good programming skills and a good background in computer networking. Some experience with distributed systems would be useful too, but not required.

Interested? Please have a look at  http://www.net.t-labs.tu-<wbr>berlin.de/~stefan/ons14ccc.pdf to get an idea of our vision, and contact us for more details.

 

Ansprechpartner: Stefan Schmid

Social network analysis: sample and pray?

Typ:  Master-Arbeit (MA)

Whenever network scientists work with real-world data -attempting for instance to analyse the spreading an opinion in a social network or to identify the source of a rumour- they have to face a common challenge: to interpret complex network-level behaviour on the basis of incomplete network connectivity. Either because of high cost or privacy, in the majority of cases the network data available for analysis are severely limited. The de-facto solution so far has been to sample and pray. Scientists work with only an incomplete knowledge of the network topology (e.g., a biased sampling), and hope that their conclusions generalize to the entire network. But can this go wrong?

This project aims to rigorously characterize the impact of sampling on existing network analysis tools. In particular, the focus will be on the graph Fourier transform (GFT), a powerful tool which has recently gained significant attention. The objective is to understand, using tools from graph theory and matrix theory, how well does GFT work when knowledge of the network (e.g., edges) is incomplete. The analytical results will be accompanied by hands on experimentation with real network datasets.

Interested? Contact us for more details, so we can find the project that best matches your skills/interests. 


Ansprechpartner: Andreas Loukas

Wireless Networking, various topics

Typ: kleines (Bachelor-) Projekt (PJ), großes (Master-) Projekt (PJ), Bachelor-Arbeit (BA), Master-Arbeit (MA), Diplomarbeit (DA)

Scope
The Berlin Open Wireless Lab (BOWL) aims towards building a wireless testbed in the area of the TUB Campus. We are using Intel XScale based hardware running the OpenWRT Linux distribution.

Audience/Requirements
The current projects in the BOWL projects are suited best to students who fancy low level programming in the Linux environment. Good knowledge of the Linux platform is required, but the specific hardware or kernel related knowledge can be aquired during the project. Knowledge of OpenWRT helps, but is not required. All software written during the project will be released as OpenSource Software.
There are always several topics available, most of them can be worked on in the scope of a project, or in a Diplom/Master Thesis. Diplom/Master thesis' will require a strong analytical and experimental evaluation of the subject, thus some skills in network measurement and statistics are required.

Topics
Topics come, among others, from the areas of

  • Node management: Monitoring, system upgrades, software distribution
  • Routing: Wireless routing protocols, client mobility
  • Transport Gateways: Proxies, redundant uplinks, Transport and Application Layer bridges
  • Service Discovery: Distributed service discovery, smart flooding, DHT-based algorithms
  • Measurement: Wireless measurement, link layer analysis, self-interference, trace synchronisation and merging

Details
Contact: Julius Schulz - Zander

ENCIX: ENCrypted Internet Exchange

Typ: kleines (Bachelor-) Projekt (PJ), großes (Master-) Projekt (PJ), Bachelor-Arbeit (BA), Master-Arbeit (MA), Diplomarbeit (DA)

Scope

Internet Exchange points provide a community driven place where network traffic can be exchanged. The whole concept follows the principle of keeping local traffic local. In this project, a student is supposed to first explore how efficient, secure and escrow free encryption of all traffic on an IXP could be performed. Specifically, this should focus on ensuring that the process is transparent to the operator of the IXP, yet does not impact inherent necessities, e.g., accounting and billing.

Details
Contact: Tobias Fiebig

Zusatzinformationen / Extras

Direktzugang

Schnellnavigation zur Seite über Nummerneingabe