sics [13, 14, 24, 31, 63]; touchless systems monitoring-sics [13, 14, 24, 31, 63]; touchless systems monitoring- sics [13, 14, 24, 31, 63]; touchless systems monitoring- tracking resource usage, ensuring system health and policy tracking resource usage, ensuring system health and policy tracking resource usage, ensuring system health and policy compliance [32, 67]; kernel integrity and security monitoring-compliance [32, 67]; kernel integrity and security monitoring- compliance [32, 67]; kernel integrity and security monitoring- intrusion detection, anti-malware, firewall and virus scan-intrusion detection, anti-malware, firewall and virus scan- intrusion detection, anti-malware, firewall and virus scan- ning [9, 23, 26-28, 34, 36, 38, 54, 64]; cloud management ning [9, 23, 26-28, 34, 36, 38, 54, 64]; cloud management ning [9, 23, 26–28, 34, 36, 38, 54, 64]; cloud management and infrastructure operations such as VM sizing and mi-and infrastructure operations such as VM sizing and mi- and infrastructure operations such as VM sizing and mi- gration, memory checkpointing and deduplication, device gration, memory checkpointing and deduplication, device gration, memory checkpointing and deduplication, device utilization monitoring, cloud-wide information flow track-utilization monitoring, cloud-wide information flow track- utilization monitoring, cloud-wide information flow track- ing and policy enforcement, cluster patch management, and ing and policy enforcement, cluster patch management, and ing and policy enforcement, cluster patch management, and VM similarity clustering [3, 8, 10, 15, 33, 58, 76].
There are different ways in which VMI gains visibility There are different ways in which VMI gains visibility There are different ways in which VMI gains visibility into the runtime state of a VM, ranging from exposing a into the runtime state of a VM, ranging from exposing a into the runtime state of a VM, ranging from exposing a raw byte-level VM memory view and traversing kernel data raw byte-level VM memory view and traversing kernel data raw byte-level VM memory view and traversing kernel data structures in it [5, 6, 14, 25, 40, 41, 46], to implanting pro- structures in it [5, 6, 14, 25, 40, 41, 46], to implanting pro- structures in it [5, 6, 14, 25, 40, 41, 46], to implanting pro- cesses or drivers into the guest [16, 29].
Therefore, applica- tion developers have different alternatives to choose from tion developers have different alternatives to choose from tion developers have different alternatives to choose from based on their desired levels of latency, frequency, over- based on their desired levels of latency, frequency, over- based on their desired levels of latency, frequency, over- head, liveness, consistency, and intrusiveness, constrained head, liveness, consistency, and intrusiveness, constrained head, liveness, consistency, and intrusiveness, constrained by their workloads, use-cases, resource budget and deploya-by their workloads, use-cases, resource budget and deploya- by their workloads, use-cases, resource budget and deploya- bility flexibility.
Discussion: (i) We do not include live snapshotting in our Discussion: (i) We do not include live snapshotting in our Discussion: (i) We do not include live snapshotting in our quantitative evaluation because of the unavailability of a quantitative evaluation because of the unavailability of a quantitative evaluation because of the unavailability of a standalone implementation (patch or library) for our KVM standalone implementation (patch or library) for our KVM standalone implementation (patch or library) for our KVM testbed, while its qualitative performance measures are bor- testbed, while its qualitative performance measures are bor- testbed, while its qualitative performance measures are bor- rowed from [35].
Another interesting observation is the markedly high im- Another interesting observation is the markedly high im- Another interesting observation is the markedly high im- pact on the disk throughputs with memory dumping, as com- pact on the disk throughputs with memory dumping, as com- pact on the disk throughputs with memory dumping, as com- pared to the CPU intensive benchmark, which moreover pared to the CPU intensive benchmark, which moreover pared to the CPU intensive benchmark, which moreover shows no improvement even when the monitoring frequency shows no improvement even when the monitoring frequency shows no improvement even when the monitoring frequency is reduced from 0.1Hz to 0.01Hz.