Niloofar Fazlollahi

Path switching and grading algorithms for advance channel reservation architectures

As a result of perceived limitations of TCP/IP in supporting high-throughput applications, significant efforts have recently been devoted to develop alternative architectures based on the concept of advance channel reservation. In this paper, we develop a polynomial-time algorithmic framework, called graded channel reservation (GCR), to support the implementation of such architectures. This framework enables users to specify minimum bandwidth and duration requirements for their connections. Upon receiving a request, GCR returns the highest graded path, selected according to a general, multicriteria optimization objective. In particular, if the optimization criterion is delay, we prove that GCR returns the earliest time available to establish the connection. Thereafter, we present a generalization of GCR, called GCRswitch, that is capable of supporting path switching throughout a connection. We present practical methods for minimizing or limiting the number of path switches. Through extensive simulations, we evaluate the performance of GCR and its variants under various topological settings and applications workload. Our results show that, for certain traffic parameters, optimized path selection combined with path switching can reduce the average delay of requests by an order of magnitude and increase the maximum sustainable load by as much as 50%.

Graded Channel Reservation with Path Switching in Ultra High Capacity Networks

We introduce a new algorithmic framework for advanced channel reservation in ultra high speed networks, called Graded Channel Reservation (GCR). GCR allows users to specify minimum bandwidth and duration requirements for their connections. GCR returns the highest graded path, selected according to a general, multi-criteria optimization objective. In particular, if the optimization criterion is delay, we prove that GCR returns the earliest time available to establish the connection. The computational complexity is polynomial in the size of the graph and the number of pending requests. We introduce a number of variants to GCR, including one that that provides the capability to switch between different paths during a connection. We present practical methods for minimizing or limiting the number of path switches. Through extensive simulations, we evaluate the performance of GCR and its variants under various topological settings and applications workload. Our results show that, for certain traffic parameters, optimized path selection combined with path switching can reduce the average delay of requests by an order of magnitude and increase the saturation throughput by as much as 50%. I.

Connected Identifying Codes

We consider the problem of generating a connected identifying code for an arbitrary graph. After a brief motivation, we show that the decision problem regarding the existence of such a code is NP-complete, and we propose a novel polynomial-time approximation ConnectID that transforms any identifying code into a connected version of at most twice the size, thus leading to an asymptotically optimal approximation bound. When the input identifying code to is robust to graph distortions, we show that the size of the resulting connected code is related to the best error-correcting code of a given minimum distance, permitting the use of known coding bounds. In addition, we show that the size of the input and output codes converge for increasing robustness, meaning that highly robust identifying codes are almost connected. Finally, we evaluate the performance ConnectID of on various random graphs. Simulations for Erdos-Rényi random graphs show that the connected codes generated are actually at most 25% larger than their unconnected counterparts, while simulations with robust input identifying codes confirm that robustness often provides connectivity for free.

Distributed advance network reservation with delay guarantees

New architectures have recently been proposed and deployed to support end-to-end advance reservation of network resources. These architectures rely on the use a centralized scheduler, which may be unpractical in large or administratively heterogeneous networks. In this work, we explore and demonstrate the feasibility of implementing distributed solutions for advance reservation. We introduce a new distributed, distance-vector algorithm, called Distributed Advance Reservation (DAR), that provably returns the earliest time possible for setting up a connection between any two nodes. Our main findings in this context are the following: (i) we prove that widest path routing and path switching (i.e, allowing a connection to switch between different paths) are necessary to guarantee earliest scheduling; (ii) we propose a novel approach for loop-free distributed widest path routing, leveraging the recently proposed DIV framework. Our routing results directly extend to on-demand QoS routing problems.

On the Capacity Limits of Advanced Channel Reservation Architectures

The next generation of grid applications demand fast and reliable transfers of extremely large volumes of data between distributed sites around the world. For example, the U.S. Department of Energys Genomes to Life (GTL) project aims at supporting critical applications such as bioenergy production or microbial carbon recycling [1]. GTL studies relies on ultra high throughput connections between research laboratories and supercomputers for processing and analyzing massive amounts of data.

Throughput-competitive advance reservation with bounded path dispersion

In response to the high throughput needs of grid and cloud computing applications, several production networks have recently started to support advance reservation of dedicated circuits. An important open problem within this context is to devise advance reservation algorithms that can provide provable throughput performance guarantees independently of the specific network topology and arrival pattern of reservation requests. In this paper, we first show that the throughput performance of greedy approaches, which return the earliest possible completion time for each incoming request, can be arbitrarily worse than optimal. Next, we introduce two new online, polynomial-time algorithms for advance reservation, called BatchAll and BatchLim. Both algorithms are shown to be throughput-optimal through the derivation of delay bounds for 1 + ε bandwidth augmented networks. The BatchLim algorithm has the advantage of returning the completion time of a connection immediately as a request is placed, but at the expense of looser delay performance than BatchAll. We then propose a simple approach that limits path dispersion, i.e., the number of parallel paths used by the algorithms, while provably bounding the maximum reduction factor in the transmission throughput. We prove that the number of paths needed to approximate any flow is quite small and never exceeds the total number of edges in the network. Through simulation for various topologies and traffic parameters, we show that the proposed algorithms achieve reasonable delay performance, even at request arrival rates close to capacity bounds, and that three to five parallel paths are sufficient to achieve near-optimal performance.

Competitive advance reservation with bounded path dispersion

Advance channel reservation is emerging as an important feature of ultra high-speed networks requiring the transfer of large files. In this paper, we present two new delay-competitive algorithms for advance reservation, called BatchAll and BatchLim. These algorithms are guaranteed to achieve optimal throughput performance, based on multi-commodity flow arguments. Unlike BatchAll, the BatchLim algorithm returns the completion time of a connection immediately as a request is placed, but at the expense of a slightly looser competitive ratio than that of BatchAll. We propose a simple approach that limits the number of parallel paths used by the algorithms while provably bounding the maximum reduction factor in the transmission throughput. We show that, although the number of different paths can be exponentially large, the actual number of paths needed to approximate the flow is quite small and proportional to the number of edges in the network. According to our simulations for a number of topologies, three to five parallel paths are sufficient to achieve close to optimal performance.

Distance Vector-based Advance Reservation with Delay Performance Guarantees

We explore and demonstrate the feasibility of implementing distributed solutions for advance reservation of network resources. We introduce a new distributed, distance-vector algorithm, called Distributed Advance Reservation (DAR), that provably returns the earliest time possible for setting up a connection between any two nodes. Our main findings are the following: (i) we prove that widest path routing and path switching (i.e, allowing a connection to switch between different paths) are necessary to guarantee earliest scheduling; (ii) we propose and analyze a novel approach for loop-free distributed widest path routing, leveraging the recently proposed DIV framework. Our routing results directly extend to on-demand and inter-domain QoS routing problems.

Connecting identifying codes and fundamental bounds

We consider the problem of generating a connected robust identifying code of a graph, by which we mean a subgraph with two properties: (i) it is connected, (ii) it is robust identifying, in the sense that the (subgraph-) induced neighborhoods of any two vertices differ by at least 2r + 1 vertices, where r is the robustness parameter. This particular formulation builds upon a rich literature on the identifying code problem but adds a property that is important for some practical networking applications. We concretely show that this modified problem is NP-complete and provide an otherwise efficient algorithm for computing it for an arbitrary graph. We demonstrate a connection between the the sizes of certain connected identifying codes and error-correcting code of a given distance. One consequence of this is that robustness leads to connectivity of identifying codes.