Thesis proposals

Full name:
Erik Eliasson
Email:
ere@kth.se
Start date:
2011-01-24
Duration in Months:
5
Location:
Background:


HDVIPER was a EU project looking into High Definition video conferencing. A solution consisting of client implementations from different companies that can connect to a virtual "room" on a server was created. The focus was high quality video conferencing with maximum flexibility. It was assumed that bandwidth limitations would not be a problem.

A key featues of the system is that every client gets all media (audio and video) from all other clients in it's highest quality. This means that in a multiparty conference, a client typically have more video information than screen realestate, and it is up to the client how to display the video streams.

This architecture is different from the most common one where a central unit decodes all video stream and combines them into one outgoing stream for each client (this design suffers from scalability problems with computational complexity and is not feasible to implement in software, and in hardware only for a limited set of participants).

On todays hardware, the HDVIPER systems provides exellent quality and flexibility up to ~5 participants. For bigger conferences there are scalability problems. The central conference server has to forward N^2 streams for N participants.

This thesis will look into how the scalability problem in the HDVIPER conference architecture can be solved using multicast.

Full name:
Prof. Bjorn Pehrson, Hervé Ntareme (PhD stude
Email:
ntareme@kth.se
Start date:
2011-01-19
Duration in Months:
5
Location:
ICT school, KTH
Background:

The Bytewalla project started with the idea to implement Delay Tolerant Networking (DTN) as an application on the Android platform. The idea is that people travelling from villages to cities and vice versa carry data on their Android phones. In the village a user download data from a WiFi access point (in a telecentre without Internet connectivity). Then, he carries the data to the city where he can connect to another WIFI access point to upload the data to the Internet. This “data-mule”operation is transparent to the user, who is able to use his mobile phone as usual. At the moment there have been 3 bytewalla software releases.


Bytewalla 3 introduces the capability to transfer the bundles (DTN data units) among android phones. It enhances flexibility, manageability, and routing of the network architecture, by incorporating two prominent features; dynamic discovery of the neighbouring nodes in the network, and PRoPHET [1].


The dynamic discovery of the neighbouring nodes enables new nodes to enter, and leave the network in a dynamic manner. It eliminates the requirement for a static allocation of host IP addresses and End point Identifiers (EID) to the network nodes as required in the earlier versions of Bytewalla. PRoPHET is an efficient routing protocol for DTN which enhances the probability, reliability and speed of message delivery.


The development of the bytewalla DTN application has been completed and the source code is open and available for download on the project's website. http://www.tslab.ssvl.kth.se/csd/projects/1031352/

Full name:
Lena Wosinska
Email:
wosinska@kth.se
Start date:
2010-01-15
Duration in Months:
5
Location:
ICT school, KTH
Background:
Today two primary signaling schemes exist for wavelength reservation in WDM optical networks; they are SIR (Source Initiated Reservation) and DIR (Destination Initiated Reservation). SIR can provide a lower setup delay but usually a higher blocking probability is incurred with this scheme mainly because of the outdated global state information. While DIR (Destination Initiated Reservation) provides with lower blocking probability however it has a higher setup delay. Also DIR schemes can suffer backward blocking because of the raceway conditions. RSVP (Resource Reservation Protocol) protocol used in GMPLS is an implementation of DIR philosophy. While CR-LDP (Constraint Routing Based Label Distribution Protocol) used in GMPLS for signaling is an example of SIR scheme.
Full name:
Lena Wosinska
Email:
wosinska@kth.se
Start date:
2010-02-01
Duration in Months:
5
Location:
ICT school, KTH
Background:

An efficient routing and wavelength assignment (RWA) algorithm and wavelength conversion capability are two primary tools for improving the blocking performance of WDM all-optical networks. Adaptive path computation is an effective way to utilize the available optical resources according to the current offered traffic and network status for the dynamic lightpath requests. Compared with the fixed alternate routing, the adaptive routing can significantly improve the blocking performance. However, resource advertisement protocols (such as link state advertisement (LSA)) must be used in adaptive routing to provide the nodes executing the RWA algorithm with the necessary information on the resources (i.e., wavelengths and wavelength conversion) availability based on the current network state. The advertisement protocol shall be designed to disseminate the timely information about the available resource (i.e., the protocol convergence time) in the network, while including the sufficient information in signaling overhead required by this task. In the conventional scheme, the advertisement mainly concerns free wavelengths on each link, their identifiers, and potentially some other transmission related parameters (e.g. wavelength conversion in wavelength convertible optical circuit switching networks). The resulting large data set of control overhead increases the complexity of finding a good compromise between the timeliness of the advertisement protocol and its signaling overhead. To address this problem, we present an opaque LSA protocol which only needs to flood and update a small data set containing the summary of the complete network status information, thus significantly reducing the amount of required control overhead. As only an information summary (IS) of the complete network status is advertised, the protocol is referred to as IS-LSA. However, it should be noted that so far most study is based on a simplistic assumption, i.e., the absence of consideration on outdated information of the link states.

Full name:
Erik Eliasson
Email:
ere@kth.se
Start date:
2011-01-24
Duration in Months:
5
Location:
Background:
The success and usability of a conferencing tool depends on it working in all or most
network environments.
Minisip is an Open Source VoIP and video conferencing platform that has basic
support for working on NATed environments by using the STUN protocol. There
are however situations where this is not enough.
Examples of such a situations where STUN is not always sufficient is if two
clients are behind the same NAT, or some types of NATs when clients are behind
two different NATs.

Full name:
Erik Eliasson
Email:
ere@kth.se
Start date:
2011-01-24
Duration in Months:
5
Location:
Background:

HDVIPER was a EU project looking into High Definition video conferencing. In the HDVIPER architecture all clients sent their video to a central forwarding server. Each client got all other video streams in highest resolution. A client tyipcally does not have enough screen realestate to display all video streams in full size at the same time. It typically scales down the videos before displaying them.

The architecture supports great flexibility and audio/video quality since no information is lost in the central server, but there are however scalability problems.

  • The central forwarding server must handle N^2 streams for conferences with N participants. This makes the central server a bandwidth bottleneck for conferences with many participants.
  • A client must support handling N-1 incoming video streams (decoding, scaling and displaying). This makes the client CPUs a computational bottleneck.

 

Information that is never presented to any user is transmitted in many scenarios, and the bottlenecks makes the architecture impractical for HD video conferences with more than ~5 participants on todays PC hardware.

 

Full name:
Prof. Bjorn Pehrson, Hervé Ntareme (PhD stude
Email:
ntareme@kth.se
Start date:
2011-01-19
Duration in Months:
5
Location:
ICT school, KTH
Background:

A Wireless Sensor Network (WSN) consists of spatially distributed autonomous sensor nodes deployed to cooperatively monitor physical or environmental conditions. In a common WSN architecture, the nodes are deployed to acquire measurements such as temperature, voltage, dissolved oxygen or other natural variables. The nodes are part of a wireless network connected to a gateway, which governs network aspects such as client authentication, clock synchronization, etc.

The gateway is one of the most important components upon which the efficiency of the sensing activity of a WSN depends. While nodes are used to report any activities happening in their surroundings and relay the information received from neighbour nodes to other nodes in transit to the gateway, the gateway collects all the information received from each node in a database and makes this information available externallyt.. Hence, the gateway provides an interface between the sensor network and the external network infrastructure and must have enough power to be able to run a database buffering the sensor data, perform necessary local calculations, read GPS position and time, distribute time in the sensor network and communicate with an external network. The gateway should, however, be low power enough to run autonomously in the field, maybe sleeping when possible and waking up by internal or external events.

A common challenge is how to connect a WSN to the Internet, especially in environments lacking network infrastructures. In this thesis, we want to explore the use of Delay Tolerant Networking Technology (DTN) to facilitate the connection of a WSN to the Internet with the usage of android smart phones. DTN is an approach to computer network architecture that seeks to address the technical issues in heterogeneous networks that may lack continuous Internet network connectivity. Examples of such networks are those operating in mobile or extreme terrestrial environments.


The Bytewalla DTN application which is a solution that has been developed at KTH/ICT/TSLab to connect rural villages to the Internet by using Android cellular phones will be used as a basis for this thesis project . The idea is that people travelling from villages to cities and vice versa carry data on their Android phones. In the village a user download data from a WIFI access point (in a telecentre without Internet connectivity). Then, he carries the data to the city where he can connect to another WIFI access point to upload the data to the Internet. This “data-mule”operation is transparent to the user, who is able to use his mobile phone as usual. The development of the Bytewalla DTN application has been completed and the source code is open and available for download on the project's website. http://www.tslab.ssvl.kth.se/csd/projects/092106/


Full name:
Karl-Johan Grinnemo
Email:
grinnemo@kth.se
Start date:
2010-01-25
Duration in Months:
6
Location:
Ericsson Research, Kista
Background:

Stream Control Transmission Protocol (SCTP), specified by Internet Engineering Task Force (IETF)
RFC 4960, is used today for niche applications, mainly for control signaling. However, due to the
number of interesting features, such as stream multiplexing, fail-over and handover mechanisms,
SCTP becomes increasingly attractive as a general purpose transport protocol for applications.

This is the practical work which aims at understanding and evaluation of various possibilities to
enable SCTP as a general purpose transport protocol for applications. One of the major
requirement is careful documentation of the results and different alternatives of using SCTP and
specifically usage of extended API to gain control over application reaction on various network
events.

Full name:
Peter Sjödin
Email:
psj@kth.se
Start date:
2010-02-01
Duration in Months:
5
Location:
ICT school, KTH
Background:

Object routing and location are techniques to perform location-independent and scalable routing, intended for large-scale distributed applications in overlay networks such as decentralized storage and distributed computations. The routing functions are performed in the end-nodes using object identifiers, for example, Tapestry uses SHA-1 signatures as identifiers. Many advantages could be gained if instead existing networking functionality could be used to perform these functions. The purpose of this project is to investigate how existing commodity IP functionality can be used for object routing, and to study possible extensions of IP for this purpose.

 

With this approach, objects will reside in the network for long periods of time. This means that the network will function both for storage and for communication. In this regard, there are two main problems in the network that deserve particulart attention:

- Object insertion The objects can be inserted in two ways:

o   Specific injections points of content where the objects are “inserted” into the network

o   Self-learning: The objects are recorded and stored in the routers when the object passes them.

 

- Object storage location: One of the main problems is that the modern storage media, such as hard disk drives, have limited throughput even if the storage size is extremely large, and that each object require a minimum I/O bandwidth at a request. A simple example is WEB-TV where the content should be play-out such that frozen viewing is avoided. The main selection must then be based on the resource models:

o   Resource model of storage media where both the size and bandwidth constraints are considered as a part of the model.

o   Constraints of the transmission resources in the model.

 

- Object forwarding: If the object is not stored in the first hop router, it has to be forwarded to the next hop. Therefore, some next-hop route-selection algorithm is required.

 

In this project, the objects are modeled as large IP-packet i.e. jumbo-frames. This means that the objects are treated as normal packets.