Georgia Fragkouli | EPFL Graph Search

I am a fifth-year PhD student in EPFL's Network Architecture Laboratory, co-advised by Prof. Katerina Argyraki and Prof. Bryan Ford. My research focuses on enhancing the trust around the performance of network systems and its implications on user anonymity and privacy of network topology. Our recent work on reconciling user anonymity with Internet transparency [ paper ] won a 2020 IETF/IRTF Applied Networking Research Prize. Prior to EPFL, I earned my diploma in Electrical and Computer Engineering from the National Technical University of Athens (NTUA).

Limiting Lamport Exposure to Distant Failures in Globally-Managed Distributed Systems

Enis Ceyhun Alp, Cristina Basescu, Pierluca Borsò, Gaylor Bosson, Kelong Cong, Bryan Alexander Ford, Georgia Fragkouli

Globalized computing infrastructures offer the convenience and elasticity of globally managed objects and services, but lack the resilience to distant failures that localized infrastructures such as private clouds provide. Providing both global management and resilience to distant failures, however, poses a fundamental problem for configuration services: How to discover a possibly migratory, strongly-consistent service/object in a globalized infrastructure without dependencies on globalized state? Limix is the first metadata configuration service that addresses this problem. With Limix, global strongly-consistent data-plane services and objects are insulated from remote gray failures by ensuring that the definitive, strongly-consistent metadata for any object is always confined to the same region as the object itself. Limix guarantees availability bounds: any user can continue accessing any strongly consistent object that matters to the user located at distance ∆ away, insulated from failures outside a small multiple of ∆. We built a Limix metadata service based on CockroachDB. Our experiments on Internet-like networks and on AWS, using realistic trace-driven workloads, show that Limix enables global management and significantly improves availability over the state-of-the-art.

2022

Toward Internet Performance Transparency

Georgia Fragkouli

From medical support to education and remote work, our everyday lives increasingly depend on Internet performance. When users experience poor performance, however, the decentralization of the Internet allows limited visibility into which network is responsible. As a result, users are promised Service Level Agreements (SLAs) they cannot verify, regulators make rules they cannot enforce, and networks with competitive performance cannot reliably showcase it to attract new customers. To change this, researchers have proposed transparency protocols, which rely on networks reporting on their own performance. However, these proposals would be hard to adopt because i) they require substantial network resources for extracting and publishing the performance information, or ii) they require cooperative networks that honestly report their performance against their self-interests, or iii) they threaten the anonymizing capability of Tor-like networks by violating their limited visibility assumptions and introducing a new attack vector against them.This dissertation enables network users to estimate the loss and delay of individual networks in an efficient and accurate manner, despite networks generating and controlling the performance data and potentially wanting to exaggerate their performance. It also proposes the first transparency protocol that tries to preserve the capabilities of anonymity networks.The key to efficient and accurate performance monitoring is i) creating incentives for networks to be honest by causing dishonest networks to get into conflict with their neighbors, and ii) combining these incentives with mathematical tools that "tie together" different aspects of network performance.The key to anonymity-preserving monitoring is the insight that users can benefit from transparency even when networks expose coarser-than-per-packet performance information, which at the same time hides sensitive communication patterns and improves anonymity.Our thesis is that efficient and accurate Internet performance transparency is possible and that we can ease the tussle between transparency and user anonymity.

EPFL2022

Destination buffer analysis for packets rejection obliteration in multi-channel networks

Georgia Fragkouli

In this paper we study a multi-channel network, each station of which is equipped with a network interface that has a receiver buffer of multiple packets. In this way, each station is able to receive multiple packets per time instant. We adopt a synchronous access protocol which is affected by the collisions over the multiple channels and the destination conflicts. The proposed protocol performance crucially depends on the size of the receiver buffer since it determines the packet rejection probability at destination. An analytical probabilistic model based on a Markovian process is adopted for the performance measures derivation by means of closed mathematical formulas. The proposed protocol is compared to relative ones that either totally ignore the receiver collisions or assume a receiver buffer of a single packet capacity. The numerical results show that the increase of the receiver buffer size improves the performance decisively, resulting in higher throughput and lower delay and rejection probability. Also, it is shown that the appropriate receiver buffer size per station in order for the rejections at destination to be effectively eliminated is not unlimited but is limited to three packets for 0.1% accepted maximum level of rejection probability. (C) 2019 The Franklin Institute. Published by Elsevier Ltd. All rights reserved.

2019

Toward Internet Performance Transparency

Georgia Fragkouli

EPFL2022