Maintaining Network QoS Across NIC Device Driver Failures Using Virtualization

Abstract

Abstract-Device driverf ailures have been shown to be a major cause of system failures. Network services stress NIC device drivers, increasing the probability of NIC driverb ugs being manifested as serverf ailures. System virtualization is increasingly used for serverconsolidation and management. The isolated driverdomain (IDD) architectureused by several virtual machine monitors, such as Xen, forms a natural foundation for making systems resilient to NIC driverf ailures. In order to realize this potential, recovery must be fast enough to maintain QoS for network services across NIC driverf ailures. Wes how that the standard Xen configuration, enhanced with simple detection and recovery mechanisms, cannot provide such QoS. However, with NIC drivers isolated in twov irtual machines, in a primary/warm-sparec onfiguration, the system can recoverf rom an overwhelming majority of NIC driverfailures in under 10ms.

Resilient Virtual Clusters

“…While some of these enhancements have been presented before [15,12,16], theya re described here to facilitate understanding of the overall system and the experimental results. The DVM recovery mechanism for block devices has not been presented in prior work.…”

“…Figure 2 shows the main components of a virtualized system with the resiliencye nhancements of the Xen VI and a virtual cluster as multiple AppVMs. The VI resiliency enhancements include ReHype [16] for detecting and recovering from VMM failure, twoD VMs to enable uninterrupted access to devices for the AppVMs, a DVM Manager for controlling recovery from DVM failure, and DVMf ailure detectors for detecting and informing AppVMs and the DVM Manager of DVM failure [15,12].…”

“…DVM crashes are detected when the crash handler in the DVM's kernel makes a hypercall to the VMM or when the VMM responds to illegalD VM activity by killing the DVM. DVM hangs are detected when the DVM stops context switching among processes or when the DVM stops consuming requests on its shared rings with AppVMs [15].…”

“…Over the last fewy ears there has been significant interest in techniques for enhancing system resiliencyt od evice driver failure [6,17,10,15,12]. These techniques isolate device drivers in a DVM and recovert he failed device driverb y rebooting the DVM or using a newd evice driveri nas pare DVM.…”

“…In particular,t his has included recovery of driver VMs (DVMs), which provide VMs safe access to shared physical I/O devices [6,17,15,12], and recovery from VMM failures [16]. However, these works focused on a single component of the VI in isolation, and their evaluations were based on microbenchmarks or other simple workload configurations.…”