Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/69646CAB-7146-4090-AEDF-1824559D19A3 | ||
| Year | 2024 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Ioannis-Iason Georgakas, "PCIe monitoring for secure code execution in heterogeneous system architectures", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2024 https://doi.org/10.26233/heallink.tuc.100243 | ||
| Appears in Collections |
Nowadays, heterogeneous systems architectures are becoming increasinglypopular for running complex and heavy-load computational tasks. Manycompanies own vast data centers consisting of thousands of heterogeneoussystems, allowing developers to run their applications in GPUs, FPGA, orTPUs. However, the rise in the use of such systems has led to a new vastattack vector with various possibilities to emerge. GPU vulnerabilities andtheir exploitations are realistic instances of such attacks. An instance of anattack on a GPU is to misuse a miner by continuously executing it and, as aresult, overload the GPU and take it out for a long time period. This thesispresents a novel method of detecting attacks in heterogeneous systems bymonitoring the PCIe traffic at a low level from the host to the correspondingPCIe endpoint. Monitoring the PCIe traffic makes it possible to check all thedata and code being transferred to an accelerator in real-time and detect ifany exhibit malicious behavior. All the PCIe traffic can be monitored basedon rules that mark behaviors or patterns as malicious. This thesis presents atool to monitor the PCIe traffic in an emulated heterogeneous system CPU-GPU using an MPSoC and an FPGA for any GPU miner running in the dis-crete GPU based on a set of rules composed of patterns presented in GPUinstructions. The rules have been developed to detect miners’ execution bychecking specific patterns of GPU instructions that indicate their presence.The monitoring tool is fully scalable and does not add processing overheadto the execution of the application. Finally, the PCIe monitoring tool achievedbetter performance than the corresponding state-of-the-art implementationin a CPU by an overall speedup in execution time of 1.55, and it is more energy efficient by 543 times the CPU’s implementation with a minimumamount of resources needed.
Copyright © DIAS 2013
