I/O processing in a virtualized platform

Abstract

Virtualization provides levels of execution isolation and service partition that are desirable in many usage scenarios, but its associated overheads are a major impediment for wide deployment of virtualized environments. While the virtualization cost depends heavily on workloads, it has been demonstrated that the overhead is much higher with I/O intensive workloads compared to those which are compute-intensive. Unfortunately, the architectural reasons behind the I/O performance overheads are not well understood. Early research in characterizing these penalties has shown that cache misses and TLB related overheads contribute to most of I/O virtualization cost. While most of these evaluations are done using measurements, in this paper we present an executiondriven simulation based analysis methodology with symbol annotation as a means of evaluating the performance of virtualized workloads. This methodology provides detailed information at the architectural level (with a focus on cache and TLB) and allows designers to evaluate potential hardware enhancements to reduce virtualization overhead. We apply this methodology to study the network I/O performance of Xen (as a case study) in a full system simulation environment, using detailed cache and TLB models to profile and characterize software and hardware hotspots. By applying symbol annotation to the instruction flow reported by the execution driven simulator we derive function level call flow information. We follow the anatomy of I/O processing in a virtualized platform for network transmit and receive scenarios and demonstrate the impact of cache scaling and TLB size scaling on performance.

Does virtualization make disk scheduling passé?

“…There has been a significant amount of investigation around optimizing processing [2][10] [18] [19], memory [9][14] [20], and I/O operations [17], though much of the work related to virtual machine I/O optimization has focused on networking [1][5] [12]. Relatively little work has been done in the area of disk I/O optimization [15].…”

“…Merging recognizes adjacent I/O requests and combines them, reducing the number of I/O requests, and associated interrupts, DMA operations, etc, that must be exchanged with the device. In a virtualized environment, it reduces the number of transitions between the guest operating system and the VMM, which are frequently the source of much of the overhead in virtualized environments [1]. Sorting arranges pending I/O requests in block order to minimize the distance that the disk heads have to move on the physical disk since seeks are the most expensive part of physical disk I/O [4].…”

Providing Performance Guarantees to Virtual Machines Using Real-Time Scheduling

“…In [5], the authors proposed a monitoring infrastructure for Xen and estimated the CPU overhead induced by I/O virtualization using a set of HTTP-based benchmarks. In [3], the authors characterized the overheads of network virtualization in Xen using full system simulation; in this way, they were able to estimate the effects of the hardware architecture on the virtualization stack performance. To increase the control over I/O virtualization, various solutions were proposed; most of them used CPU scheduling to isolate the VMs from the performance perspective.…”

A Simulation Framework for the Analysis of the TLB Behavior in Virtualized Environments

“…Chadha et al [11] have used functional models to study the TLB behavior of I/O-intensive virtualized workloads. Tickoo et al [12] and Venkatasubramanian et al [13] have investigated TLB tagging with domain 1 -specific and processspecific tags.…”