Assessing performance overhead of Virtual Machine Introspection and its suitability for malware analysis
Paakkola, Sebastian (2020-06-25)
Assessing performance overhead of Virtual Machine Introspection and its suitability for malware analysis
(25.06.2020)
Julkaisu on tekijänoikeussäännösten alainen. Teosta voi lukea ja tulostaa henkilökohtaista käyttöä varten. Käyttö kaupallisiin tarkoituksiin on kielletty.
avoin
Julkaisun pysyvä osoite on:
https://urn.fi/URN:NBN:fi-fe2020081961058
https://urn.fi/URN:NBN:fi-fe2020081961058
Tiivistelmä
Virtual Machine Introspection is the process of introspecting guest VM’s memory and reconstructing the state of the guest operating system. Due to its isolation, stealth and full visibility of the monitored target, VMI lends itself well for security monitoring and malware analysis. The topics covered in this thesis include operating system and hypervisor concepts, the semantic gap issue, VMI techniques and implementations, applying VMI for malware analysis, and analysis of the performance overhead.
The behaviour and magnitude of the performance overhead associated with doing virtual machine introspection is analysed with five different empirical test cases. The intention of the tests is to estimate the costs of a single trapped event, determine the feasibility of various monitoring sensors from usability and stealth perspective, and analyse the behaviour of performance overhead.
Various VMI-based tools were considered for the measurement, but DRAKVUF was chosen as it is the most advanced tool available. The test cases go as follows. The chosen load is first executed without any monitoring to determine the baseline execution time. Then a DRAKVUF monitoring plugin is turned on and the load is executed again. After both measurements have been made, the difference between the two execution times is the time spent executing monitoring code. The execution overhead is then determined by calculating the difference between the two execution times and dividing it by the baseline execution time.
The disc consumption and execution overhead of a sensor, which captures removed files is small enough to be deployed as a monitoring solution. The performance overhead of system call monitoring sensor is dependant on the number of issued system calls. Loads which issue large numbers of system calls cause high performance overhead. The performance overhead of such loads can be limited by monitoring a subset of all system calls.
The behaviour and magnitude of the performance overhead associated with doing virtual machine introspection is analysed with five different empirical test cases. The intention of the tests is to estimate the costs of a single trapped event, determine the feasibility of various monitoring sensors from usability and stealth perspective, and analyse the behaviour of performance overhead.
Various VMI-based tools were considered for the measurement, but DRAKVUF was chosen as it is the most advanced tool available. The test cases go as follows. The chosen load is first executed without any monitoring to determine the baseline execution time. Then a DRAKVUF monitoring plugin is turned on and the load is executed again. After both measurements have been made, the difference between the two execution times is the time spent executing monitoring code. The execution overhead is then determined by calculating the difference between the two execution times and dividing it by the baseline execution time.
The disc consumption and execution overhead of a sensor, which captures removed files is small enough to be deployed as a monitoring solution. The performance overhead of system call monitoring sensor is dependant on the number of issued system calls. Loads which issue large numbers of system calls cause high performance overhead. The performance overhead of such loads can be limited by monitoring a subset of all system calls.