| Cloud Computing is a computing paradigm that has emerged in the past few years as a very promising way of using highly scalable and adaptable computing resources, as well as accessing them from anywhere in the world and from any terminal (mobile phone, tablet, laptop...). It allows companies or individuals to use computing infrastructures without having to physically own them, and without the burden of maintenance, installations, or updates. To achieve that, Cloud Computing uses already known and tested technologies such as virtualization and web services. Several Cloud Computing models exist, divided by how much of the infrastructure the user is in charge of, ranking from nothing at all (the provider is in charge of everything, from the operating system to the applications installed) to managing a whole virtual machine without even an operating system preinstalled. For instance, sharing and accessing a document on Dropbox, or running a resource intensive application on a rented machine, are both examples of what can be done with Cloud Computing. In the case of Dropbox, the user doesn't care what resources are allocated for his requests, no more than he needs to know on what operating system the request was run or how the database was accessed. But all those aspects will be part of what the user has to know and adjust in the second case. A Cloud Computing network can be public, as it is the case for Amazon, which allows you to access its resources if you pay for it, or private, if a company decides to build a cluster for its own needs. The strong appeal of Cloud Computing, for both businesses and individuals, dramatically increases the security risks, because it becomes a key target for attackers. This increased risk, added to the confidence the users must have in their services provider when it comes to managing and protecting their data, may explain why many are still reluctant to take the leap to Cloud Computing. For instance, a company may be reluctant for confidentiality reasons, while individuals may hesitate for privacy concerns. The broad range of technologies used in Cloud Computing makes it at risk for a wide variety of attacks, since it already comes with all the vulnerabilities associated with any conventional network, and all the security breaches that affect virtual machines. However, those threats are usually well documented and easily prevented. But this is not the case of the vulnerabilities that come from the use of web services, that are heavily used in Cloud Computing.;Our first contribution is to emphasize the vulnerable equipments in a Cloud Computing network, and the various ways to attack them, as well as the various forms an XML-DoS attack can take. Our second contribution is then to use, modify and improve a Cloud Computing simulator (the GreenCloud simulator, based on NS2), in order to make the study of XML-DoS attacks possible. Once the changes are made, we show the efficiency of XML-DoS attacks, and the impact they have on legitimate users. In addition, we compare the main existing defenses against XML-DoS attacks and web services attacks in general, and pick the one that seems to be best suited to protect Cloud Computing networks. We then put this defense to the test in our simulator, to evaluate its efficiency. This evaluation must take into consideration not only the ability to mitigate the attack we led in the previous step, but also the number of false positives and false negatives. One of the major challenges is to have a defense that will conciliate the ability to protect all the machines in the network, while still being able to adapt to the great heterogeneity of the various web services hosted at the same time in a Cloud Computing network. Those experimentations are then subjected to discussions and conclusions on the decisions to take when it comes to XML-DoS attacks. In particular, what defenses should be adopted and what practices should be followed, because the evaluation of the defense at the previous step will show that it may not be the optimal solution; this will be our final contribution. We made the assumption that all the Cloud Computing models could be the target of a XML-DoS attack in some way. No matter the model, it can actually use web services and is then vulnerable to those attacks, whether it is a web server handling incoming requests for all the users, or a web server a user installed on the virtual machine he rents. We thought it was essential to take into consideration the specificity of virtual machines, such as the contention for resources when they are located on the same physical machine.;Cloud Computing networks aim at being accessible from all over the world and on almost any device, and that implies using web services. Yet, web services are extremely vulnerable to XML-DoS type of attack. Those attacks take advantage of Simple Object Access Protocol (SOAP) requests using a malicious XML content. Those requests can easily deplete a web server from its resources, be it CPU or memory, making it unavailable for legitimate users; this is exactly the goal of a denial-of-service attack. The XML-DoS attacks are extremely interesting in two ways. First, they are very hard to detect, since the attack takes place on the application layer, so the user appears to be legitimate (it's impossible to detect it on the TCP/IP layer). Second, the resources needed to mount the attack are very low compared to what is needed on the web server side to process even a basic but malformed request. This type of attack has been surprisingly left quite aside, despite its efficiency and the omnipresence of web services in Cloud Computing networks. This is the reason why we decided to prove and quantify the impact such attacks can have on Cloud Computing networks, to later propose possible solutions and defenses. We estimated that using a simulated environment was the best option for various reasons, like the possibility to monitor the resources of all the servers in the network, and the greater freedom we had to build our own topology. |
