Acceleration Of TLS Protocol And Design Of Corresponding Heterogeneous Computation Framework Based On Integrated GPGPU

Along with fast development of networking technologies,the speed of Internet connections becomes faster and faster,and grows exponentially.How to transmit such large amount of data,including many different types of private information,becomes a challenging problem,and it affects everyone using the Internet,The Transport Layer Security(TLS)protocol provides security to the data of transport layer,and has been deployed widely.However,due to the high demand on services with private informa-tion,the increase of cryptography algorithm complexity,and the growth of users in the Internet,how to speed up TLS transactions processing with limited hardware resources and reduce the delay to improve user experience,have become an important problem that cannot be avoided.In order to solve this problem,many types of special computing devices(e.g.ASIC,FPGA,Discrete GPGPU)are used in the system of TLS protocol for acceler-ation.However,the study on cost-efficient integrated GPGPU is still open.Compared with commonly used hardware,integrated GPGPU has special architecture which uses shared memory and last level cache with CPU.Additionally,it has low power and price,and is easy to be developed with OpenCL.On a heterogeneous computing platform,it is interesting to study how to design a framework,which can use the capacity of different types of hardware devices by distributing jobs with high computing complexity,and can be extended to include new devices in the future.Besides,every specific algorithm can be optimized respectively in this framework.Under the background where network security becomes more and more important and the hardware devices have more and more types,on the integrated GPGPU plat-form,with the help of OpenCL,this thesis designs a heterogeneous computing frame-work with extendibility for TLS protocol.It includes our implementations of specific algorithms,which applies our improvements and optimizing strategies,in order to speed up TLS processing and decrease delay.The main contributions and innovations of this thesis can be summarized as follows:1.Exploration to the capacity of integrated GPGPU under the background of net-work security.Compared with discrete GPGPU with high performance,inte-grated one could almost eliminate the overhead of data transmission between GPU and CPU,benefitting from specific architecture.With the example of AES algorithm,this thesis makes a comparison between integrated GPGPU and CPU in executing performance.According to the experimental results,we conclude that integrated GPGPU is fit for AES encryption acceleration,but cannot bring gains in AES decryption.Then,by the comparison between integrated GPGPU and discrete GPGPU,we show the advantages of integrated one in the power and floating point computing capacity scale.Our exploration fills the gap of applying integrated GPGPU to the general purpose computation under the background of network security.2.Improvements and optimizations to the parallel RSA algorithm.This thesis ana-lyzes the RSA decryption algorithm which is widely used in building the secure connections of TLS protocol,and proposes improved parallel RSA algorithm which reduces synchronizations to only once.On the one hand,the reduction of synchronizations decreases the time for I/O and waiting significantly.On the other hand,it increases the mission complexity between two synchronizations,which provides enough space for optimizations.Thus,aiming at the properties of integrated GPGPU,we propose several optimizing strategies to take full ad-vantage of its capacity.According to the experimental results,we show that our proposed scheme can fully utilize the capacity of integrated GPGPU,and 2 times throughput improvement over the CPU only solution can be expected.3.Design of heterogeneous computing framework and mission allocating strategies.Inspired by the thread pool technologies,we design a heterogeneous computing framework with extandibility for TLS protocol.Besides,we propose a strategy of mission distribution to be applied to the system,in order for the capacity of different types of computing devices to be simply and efficiently utilized for ac-celeration.On the Mini PC platform which is equiped with CPU and integrated GPGPU,we verified the effectiveness of our designed framework and strategy by experiments.According to the experimental results,the application of integrated GPGPU can speed up the building of secure connection and the transmission of data from upper layer.It also shows the advantages of integrated GPGPU under the circumstance of limited space and power.