Βασικό ζητούμενο στον τομέα των δικτύων υψηλής ταχύτητας είναι αυτό της “διατήρησης της διαπερατότητας”. Η προσπάθεια επίτευξης του ζητούμενου αυτού στην περίπτωση παραδοσιακών δικτύων που χρησιμοποιούν προτυποποιημένα πρωτόκολλα μπορεί να γίνει μόνο στην κατεύθυνση του σχεδιασμού και της υλοποίησης του συστήματος επεξεργασίας που εκτελεί τα πρωτόκολλα αφού οι μηχανισμοί και η σύνταξη των ίδιων των πρωτοκόλλων είναι ήδη καθορισμένα. Στην παρούσα εργασία διερευνώνται τρόποι επίτευξης του ζητούμενου αυτού για το πιο βασικό από τα δικτυακά υποσυστήματα, του προσαρμογέα δικτύου. Οι σχεδιαστικές αποφάσεις που αφορούν την αρχιτεκτονική ενός network adapter παίζουν καταλυτικό ρόλο στην απόδοσή του. Στις επόμενες παραγράφους θα παρουσιαστεί η αρχιτεκτονική ενός απλού δομικά αλλά και στην σύλληψή του προσαρμογέα και στην συνέχεια θα διερευνηθούν διάφορες αρχιτεκτονικές παρεμβάσεις που στόχο θα έχουν την βελτίωση της απόδοσης αυτού. Με την βοήθεια της βιβλιοθήκης εξομοίωσης CSIM της Mesquite οι αρχιτεκτονικές που θα προκύψουν μοντελοποιούνται και δίνοντας στα χαρακτηριστικά τους τιμές προϊόντων που διατίθενται αυτή την στιγμή στην αγορά υπολογίζεται η βελτίωση που επιφέρουν. / One of the main challenges in high-throughput networking is the “throughput preservation problem”. The effort to address this problem in the case of standardized protocols should be focused in the design and implementation of the protocol processing systems, for the mechanisms and syntax are already well defined in a standardized protocol. This paper explores the ways of preserving the throughput of network adapters, the most basic of communication systems. Design and architectural issues of network adapters affect their performance characteristics. In the following sections the architecture of a simple structurally and conceptually network adapter will be presented. Next, several architectural modifications will be imposed in order to ameliorate its performance. By using Mesquite's CSIM modeling and simulating tool, models of the produced network adapter configurations will be implemented and their performance will be measured. These measurements will be produced by giving in the models' characteristics values of currently available in market products.

Abstract In the contemporary world of information technology you find a multitude of applications and systems covering a broad spectrum of areas of need in different companies. One effect of this multitude of programs is the difficulty to make them exchange information with each other or to collaborate, since they are developed by different programming languages for different platforms, with different standards and different data formats. Our aim with this work is to describe how it is possible to tie these programs together to make them actually communicate with each other in order to exchange information, share their native methods and also to become a part of the overall business processes. In this integration task you will, among other things, find different levels of application integration such as data level, method level, application interface level and user interface level integration. Application integration also involves hardware components, called middleware, that facilitate the physical connection between applications. There is a range of different middleware products offered today on the market. The functionalities of those products varies greatly depending on which technology or technologies they are built around and which vendor they come from. In order to make a connection to a real life situation we have made up a company with a need of integration. By trying to choose a solution for this company we discuss what to integrate, data, methods etc, and the technical solution, middleware product, which might be useful to integrate the different kind of applications in our imaginary company. Finally we have come up with some conclusions in our work. These are both of a more general art and also a conclusion specific to our case study.

This thesis investigates Parameter-Efficient Fine-Tuning (PEFT) methods, specifically Low-Rank Adaptation (LoRA) (Hu et al. 2021) and Adapters (Houlsby et al. 2019), using the General Language Understanding Evaluation (GLUE) dataset (Wang et al. 2019). The primary focus is to evaluate the effectiveness and efficiency of these methods in fine-tuning pre-trained language models. Additionally, we introduce a novel application by applying the methodology from Yang et al. 2024 to the adapter module weights. We utilize Laplace approximations over both the LoRA (Yang et al. 2024, Daxberger et al. 2022a) and the newly adapted Adapter weights, assessing the Expected Calibration Error (ECE) and Negative Log-Likelihood (NLL). Furthermore, we discuss practical considerations such as training time, memory usage, and storage space implications of these PEFT techniques. The findings provide valuable insights into the trade-offs and benefits of using LoRA and Adapters for fine-tuning in resource-constrained environments.

Η μικρή ταχύτητα επεξεργασίας των πακέτων που μεταδίδονται στα δίκτυα σε σχέση με την ταχύτητα μετάδοσης τους μέσα σε αυτά, δημιουργεί την ανάγκη για την εφαρμογή καινοτομιών στα συστήματα δικτύωσης, με σκοπό την ελάττωση αυτού του χάσματος και την καλύτερη εκμετάλλευση των μεγάλων ταχυτήτων μετάδοσης δεδομένων. Το πρόβλημα αυτό είναι γνωστό ως «πρόβλημα διατήρησης της ρυθμαπόδοσης» Η ενσωμάτωση επεξεργαστών στα δικτυακά συστήματα έχει βοηθήσει στην αντιμετώπιση του προβλήματος. Μια αρχιτεκτονική που προτείνεται για αυτούς τους επεξεργαστές πρωτοκόλλου, όπως ονομάζονται, εισάγει τη χρήση μιας καινοτόμας δομής καταχωρητών με την ονομασία Τripod. Η ιδέα της είναι η αντικατάσταση του επεξεργαστή που βρίσκεται στο εσωτερικό ενός προσαρμογέα δικτύου, από έναν επεξεργαστικό πυρήνα και τρεις ξεχωριστές πανομοιότυπες ομάδες καταχωρητών. Το όλο σύστημα θα λειτουργεί σε μια λογική διοχέτευσης (pipeline) με τα εξής στάδια: φόρτωσης, επεξεργασίας, εκφόρτωσης. Σκοπός αυτής της διπλωματικής είναι η σχεδίαση ενός υποσυτήματος το οποίο θα διαχειρίζεται αυτές τις ομάδες καταχωρητών και θα επιτρέπει στο σύστημα να λειτουργεί σύμφωνα με τις προδιαγραφές. Πιο συγκεκριμένα, θα υλοποιεί την φόρτωση και εκφόρτωση δεδομένων προς και από τους καταχωρητές, καθώς και την σύνδεση της κατάλληλης ομάδας καταχωρητών με τον επεξεργαστικό πυρήνα. / The low processing speed of packets that are transmitted in the networks compared to their transmission speed, creates the need for inserting innovations in the network systems, aiming at the alleviation of this gap and the better exploitation of the high transmission speeds of data. This problem is known as “the throughput preservation problem” The incorporation of embedded processors in the network systems has helped in the confrontation of the problem. An architecture that is proposed for these protocol processors, as they are named, imports the use of an innovative register structure, called Tripod. The idea is the replacement of the processor that is found in the interior of a network adapter, with a processor core and three separate similar register files. The system will function in a logic of pipeline with the following stages: loading, processing, unloading. Aim of this diploma thesis is the designing of a subsystem that will manage these register files and make the system to function according to the specifications. More concretely, it will execute the loading and unloading of data to and from the registers, as well as the connection of the suitable register file with the processor core.

Wir betrachten offene Systeme; Systeme mit Schnittstelle und Verhalten in Form eines Protokolls. Offener Systeme interagieren durch den Austausch von Nachrichten. Die Interaktion ist korrekt, wenn die interagierende offenen Systeme terminieren können. Die korrekte Interaktion unabhängig entwickelter, offener Systeme ist nicht immer möglich. Obwohl ein offenes Systeme Funktionalität bereitstellt, die ein zweites offenes System benötigt, können Inkompatibilitäten in den Schnittstellen oder im Verhalten dazu führen, dass Nachrichten nicht korrekt ausgetauscht werden. Hier kommt die Idee eines Adapters zum Tragen, der solche Unterschiede ausgleicht. Wir führen eine Technik ein, die Inkompatibilitäten der Schnittstelle und des Verhaltens getrennt voneinander betrachtet. Mit Hilfe von Transformationsregeln geben wir an, wie Nachrichten zweier offener Systeme im Zusammenhang stehen. Wir modellieren so korrekten Datenfluss zwischen den Schnittstellen der offenen Systeme. Mit Hilfe existierender Techniken zur Controllersynthese erzeugen wir korrekten Kontrollfluss, sodass die Transformationsregeln so angewendet werden, dass Inkompatibilitäten im Verhalten überwunden werden. Wir erzeugen korrekten Kontrollfluss. Die Einheit aus Transformationsregeln und Controller bezeichnen wir als Adapter. Aufbauend auf dieser Technik betrachten wir in dieser Arbeit folgende Fragestellungen: - Welche Eigenschaften besitzt die vorgestellte Technik? Wann existiert ein Adapter? Wie lässt sich die Technik allgemein auf offene, kommunizierende Systeme anwenden? - Wie verteilen wir einen generierten Adapter auf verschiedene Komponenten? - Welche Informationen können wir bereitstellen, falls wir keinen Adapter finden? - Wie leiten wir ein formales Modell eines Adapters ab, wenn wir nur aufgezeichnetes Verhalten der offenen Systeme, aber keine formalen Modelle gegeben haben? / We consider open systems being described by an interface and behavior in form of an protocol. Interaction between open systems means exchange of messages. For the interaction to be correct, the interacting open systems must be able to terminate. The interaction of independently developed open systems is not guaranteed to be correct. Although one open system provides functionality a second open system requires, incompatiblities of the interfaces or the behavior may prevent the correct exchange of messages. The idea is to introduce an adapter to overcome the incompatiblities. We present a technique that considers incompatiblities of the interfaces and the behavior separately. With transformation rules we specify how messages of two open systems relate to each other. Using these rules we model the data flow between open systems and overcome incompatiblities of the interfaces. By using existing techniques for controller synthesis we overcome incompatiblities of the behavior. Thus we create correct control flow. We call the composition of transformation rules and controller an adapter. Based on the technique presented in this thesis, we want to answer the following reseach questions: - What are the properties of the proposed technique? When does an adapter exist? How can we apply the technique on open, communicating systems in general? - How do we distribute a generated adapter to different components? - Which information can we provide, if adapter synthesis fails? - How can we discover a formal model of an adapter, when we only have recorded behavior of open systems, but no formal models?

Introduction: The use of high frequency oscillatory ventilation is increasing in treatment ofacute respiratory distress syndrome over the past decade. The technique of HFOV of ventilatingthe lungs at volumes less than the anatomical dead space calms the clinical concerns surroundingventilating stiff ARDS lungs with high pressures and volumes. This largely reduces theprobability of barotraumas and/or atelectrauma. Purpose: The study was on an in vitro bench model that answered the following researchquestions: 1. The effect of three inline closed suction adapters on delivered tidal volume duringHFOV with varying lung compliance 2. The effect of varying compliance on the amplitudedelivered by HFOV; and 3. The effect of compliance on tidal volume delivered by HFOV. Method: An in vitro bench model using high fidelity breathing simulator (ASL 5000, IngMarMedical) simulating an adult patient with ARDS was set up with 3100B SensorMedic highfrequency ventilator. The simulation included varying the compliance for each lung at 50, 40, 30and 20cmH2O while maintaining fixed resistance of 15 cmH2O/L/sec. The ventilator was set tothe following parameters: power of 6, frequency (f) of 5, inspiratory time (Ti) of 33%, bias flow(BF) of 30 LPM and oxygen concentration of 50%. The breathing simulator was connected withthe high frequency ventilator using a standard HFOV circuit and a size 8.0mm of endotrachealtube. Fourteen French Kimberly Clark suction catheters (with T and Elbow adapters) and Air-Life suction catheters (Y adapter) were placed in-line with the circuit successively to carry outthe study. Each run lasted for 1 minute after achieving stable state conditions. Thisapproximated to 300 breaths. The data was collected from the stimulator and stored by the hostcomputer. Data Analysis: The data was analyzed using SPSS v.11 to determine the statistical significance.A probability value (P value) of ≤ 0.001 was considered to be statistically significant. Results: The data analysis showed that Air-Life Y-adapter suction catheters caused the least lostin tidal volume when placed in line with HFOV and hence proved to be the most efficient. Thestudy also showed a direct relationship between amplitude and lung compliance i.e. an increasein lung compliance caused an associated increase in amplitude (power setting remainingunaltered). Lastly, the study did not show a statistically significant change in tidal volume withchanges in lung compliance. Future studies may be required to further evaluate the clinicalsignificance of the same. Conclusion:1. Many factors affect delivery of tidal volume during high frequency ventilation and thus it isnot constant. Choice of in-line suction system to be placed in line is one of the determinants ofthe same.2. Lung compliance changes lead to associated changes in amplitude delivery by HFOV. Thisshould be adjusted as patient condition improves by altering the power settings to ensure optimalventilation and to avoid trauma to the lungs.

Les services Web ont émergé comme un support de développement et d’intégration d’applications ou de systèmes d’information. Dans ce cadre, les interactions entre deux applications consommateur et fournisseur, encapsulées par des services Web se font par échanges de messages. Ces échanges s’appuient sur la notion d’interfaces, qui décrivent les interactions dans lesquelles un service peut s’engager et les dépendances entre ces interactions. Dans le Web actuel, il arrive très fréquemment que de nombreux services répondent à un même ensemble de besoins fonctionnels. Ces services sont souvent offerts par le biais d’interfaces différentes. Des nombreuses raisons, telles que la panne du service fournisseur, peuvent amener un consommateur à substituer son fournisseur habituel par un autre fournisseur qui offre la même fonctionnalité. Cette substitution provoque des incompatibilités entre l’interface du service consommateur et celle du service fournisseur substitut. Cela est dû au fait que le service consommateur n’a pas été fait en fonction de ce nouveau service fournisseur. Les recherches que nous menons dans cette thèse visent à résoudre le problème des incompatibilités dans des interactions entre deux services consommateur et fournisseur substitut. En particulier, notre contribution s’étend, tant sur le plan théorique que sur le plan pratique. Il s’agit d’une part d’un canevas pour la génération automatique des adaptateurs des interactions entre deux services. D’autre part, nous proposons une architecture logicielle multicouche fournissant un cadre permettant une substitution transparente et flexible d’un service fournisseur par un autre service vis-à-vis du consommateur du premier. Dans notre canevas pour la génération automatique des adaptateurs, une modélisation des interfaces de services en des automates a été adoptée. Puis, une étape de détection des incompatibilités entre ceux-ci est réalisée. Ensuite, un adaptateur des interactions entre les deux services est généré automatiquement sur la base d’incompatibilités détectées. La génération de l’adaptateur est guidée par le modèle d’automates. Cela permet de modéliser l’adaptateur indépendamment de son implémentation cible, permettant ainsi une bonne réutilisation des modèles. Une fois généré, l’automate de l’adaptateur suffisamment détaillé est projeté sur la technologie CEP (Complexe Event Processing). Cette projection est réalisée à l’aide des composants cartouches (en anglais : Templates) que nous avons mis en oeuvre. Chaque cartouche étant conçue pour générer du code exécutable (en termes de requêtes continues) pour la technologie CEP / Web services have emerged as a support for development and integration of applications and information systems. In this context, the interactions between two consumer and supplier applications, encapsulated by Web services are done by exchanging messages. These exchanges are based on the concept of interfaces, which describe the interactions in which a service can handle and dependencies between these interactions. In the current Web, it is very often that many services meet the same set of functional requirements. These services are often delivered through different interfaces. For many reasons, such as failure of the service provider, the consumer has to replace his usual supplier with another supplier that offers the same functionality. This substitution leads to incompatibilities between the interfaces of the service consumer and new service provider. This is because the customer service was not been developed according to the new service provider. The research we conduct in this thesis aims to solve the problem of incompatibility in the interaction between two services; consumer and new provider. In particular, our contribution extends both in theory and in practice. It is a part of a framework for the automatic generation of adapters for interactions between two services. On the other hand, we propose a multi-layer software architecture providing a framework for transparent and flexible substitution of a service provider by another with respect to an existed consumer. In our framework for automatic generation of adapters, services interfaces modeling using automata has been adopted. Then, a step of detecting incompatibilities between them is achieved. After that, an adapter of the interactions between the two services is generated automatically based on the detected incompatibilities. The generation of the adapter is based on the automata model. The generated adapter automaton contains a sufficient detail for the projected technology CEP (Complex Event Processing). This projection is performed using components templates that are implemented by us. Each template is designed to generate executable code (in terms of continuous requests) for the CEP technology