Yerevan, Yerevan 0014
Abstract of Addressing Security Issues in Cloud Computing: Cloud computing raises some very novel security issues. In the outsourced data setting users store data on the cloud and want to be able to read and modify it using an expressive language such as SQL to decide which data is accessed. Preserving data security, and specially confidentiality, in such a setting is relatively complex, specially if access patterns must be hidden too. In the outsourced computation setting a user sends some data to the cloud so that some computation is done over the data and an answer returned. Again, keeping data private is hard as traditional encryption cannot be used as it would prevent doing the computation on the cloud. In both settings, other novel security issues appear. In the outsourced data setting, how can we allow some people, unknown at the moment of outsourcing the data, to have access to a subpart of the data? Is it possible to give access to arbitrary functions (instead of only a subset) of the outsourced data? In the outsourced computation data, can the service provider certify that he has done the requested computation he was paid for and that the returned value is the result of that computation? In this presentation we will first describe briefly these security issues. We will then present in detail the main tool to solve them, homomorphic encryption, and more succinctly variants and specific protocols (functional encryption, certified computation, Oblivious RAM, Private Information Retrievel, etc.).
Abstract of Developping high performance cryptographic applications: Since the mid nineties using integer lattices (vectors of integers and their linear combinations with integer coefficients) to build cryptographic systems has brought groundbreaking revolutions to cryptography. In the nineties it brought worst-case to average-case reductions (if an attacker can break random instances with non negligible probability then he can break any instance with overwhelming probability). In the noughties it brought fully homomorphic encryption (it is possible to evaluate arbitrary circuits over encrypted data). These last years it has proved that implemented correctly it can bring extremely high performance results. In this presentation we will describe ideal lattices, and how to represent lattice elements to have extremely low computational costs. As an example we present an implementation (under submission) of a standard public key encryption scheme that is able to produce ciphertexts at or decrypt them at more than 5Gbits/s on a commodity CPU (which is comparable to AES). Moreover, in order to increase the security bits linearly, computational costs only increase linearly (as a comparison in RSA costs increase as a degree 7 polynomial), as well as plaintext size which means that the 5Gbits/s throughput performance is mainted even in the highest security settings. In fact we show that the main performance limitation is randomness generation (which also limits maximum security to 256 in our implementation).
Presenter: Dr. Carlos Aguilar (INP Toulouse, France) is a former student of the Ecole Polytechnique of Paris, Carlos Aguilar Melchor obtained a PhD in computer science in 2006 under the supervision of Yves Deswarte. He is Associate professor since 2009. His research domains are security, cryptography and privacy protection.
Please be aware that the number of participants is limited.
The tutorial is orginized within the INARMERA-ICT (Integrating Armenia into ERA: Information and and Communication Technologies) Project funded by the EU Seventh Framework Programme.
Abstract of Addressing Security Issues in Cloud Computing: Cloud computing raises some very novel security issues. In the outsourced data setting users store data on the cloud and want to be able to read and modify it using an expressive language such as SQL to decide which data is accessed. Preserving data security, and specially confidentiality, in such a setting is relatively complex, specially if access patterns must be hidden too. In the outsourced computation setting a user sends some data to the cloud so that some computation is done over the data and an answer returned. Again, keeping data private is hard as traditional encryption cannot be used as it would prevent doing the computation on the cloud. In both settings, other novel security issues appear. In the outsourced data setting, how can we allow some people, unknown at the moment of outsourcing the data, to have access to a subpart of the data? Is it possible to give access to arbitrary functions (instead of only a subset) of the outsourced data? In the outsourced computation data, can the service provider certify that he has done the requested computation he was paid for and that the returned value is the result of that computation? In this presentation we will first describe briefly these security issues. We will then present in detail the main tool to solve them, homomorphic encryption, and more succinctly variants and specific protocols (functional encryption, certified computation, Oblivious RAM, Private Information Retrievel, etc.).
Abstract of Developping high performance cryptographic applications: Since the mid nineties using integer lattices (vectors of integers and their linear combinations with integer coefficients) to build cryptographic systems has brought groundbreaking revolutions to cryptography. In the nineties it brought worst-case to average-case reductions (if an attacker can break random instances with non negligible probability then he can break any instance with overwhelming probability). In the noughties it brought fully homomorphic encryption (it is possible to evaluate arbitrary circuits over encrypted data). These last years it has proved that implemented correctly it can bring extremely high performance results. In this presentation we will describe ideal lattices, and how to represent lattice elements to have extremely low computational costs. As an example we present an implementation (under submission) of a standard public key encryption scheme that is able to produce ciphertexts at or decrypt them at more than 5Gbits/s on a commodity CPU (which is comparable to AES). Moreover, in order to increase the security bits linearly, computational costs only increase linearly (as a comparison in RSA costs increase as a degree 7 polynomial), as well as plaintext size which means that the 5Gbits/s throughput performance is mainted even in the highest security settings. In fact we show that the main performance limitation is randomness generation (which also limits maximum security to 256 in our implementation).
Presenter: Dr. Carlos Aguilar (INP Toulouse, France) is a former student of the Ecole Polytechnique of Paris, Carlos Aguilar Melchor obtained a PhD in computer science in 2006 under the supervision of Yves Deswarte. He is Associate professor since 2009. His research domains are security, cryptography and privacy protection.
Please be aware that the number of participants is limited.
The tutorial is orginized within the INARMERA-ICT (Integrating Armenia into ERA: Information and and Communication Technologies) Project funded by the EU Seventh Framework Programme.