Maquinas virtuales un nuevo mundo para medir el rendimiento

Virtual Machines: A Whole New World For Performance Analysis
Stanislav Bratanov Roman Belenov
Intel Corporation Nizhniy Novgorod, Russia

Nikita Manovich

{stanislav.bratanov, roman.belenov, nikita.manovich}@intel.com

ABSTRACT
This article addresses a problem of performance monitoring inside virtual machines (VMs). It advocates focused monitoring of particular virtualized programs,explains the need for and the importance of such an approach to performance monitoring in virtualized execution environments, and emphasizes its benefits for virtual machine manufacturers, virtual machine users (mostly, software developers) and hardware (processor) manufacturers. The article defines the problem of in-VM performance monitoring as the ability to employ modern methods and hardwareperformance monitoring capabilities inside virtual machines to an extent comparable with what is being done in real environments. Unfortunately, there are numerous reasons preventing us from achieving such an ambitious goal, one of those reasons being the lack of support from virtualization engines; that is why a novel method of ‘cooperative’ performance data collection is disclosed. The method impliescollection of performance data at physical hardware and simultaneous tracking of software states inside a virtual machine. Each statistically visible execution point of the virtualized software may then be associated with information on real hardware events. The method effectively enables time-based sampling of virtualized workloads combined with hardware event counting, is applicable tounmodified, commercially available virtual machines, and has competitive precision and overhead. The practical significance and value of the method are further illustrated by studying a parallel workload and uncovering virtualization-specific performance issues of multithreaded programs.

General Terms
Algorithms, Performance, Measurement

Keywords
Virtual machines, hardware performance eventcounters

1. INTRODUCTION
Many modern processors are going to have multiple cores and built-in virtualization technology from now on, and that clearly defines future computations to be more powerful, more parallel, and performed at a higher level of security, isolation, and even abstraction. Virtual machines (VMs) are a crucial piece of technology that binds those properties together, paying the priceof complexity and, to some extent, obscurity. But the greater the complexity the bigger the desire to understand and the greater the need for analysis, debugging and performance tuning instruments. Not only to tune virtualization engines themselves – but rather to provide the same level of support to virtual machine users (in a general sense, that is, at least software developers and virtualmachine manufacturers), who might not even be aware of their being virtualized, as if operating on real hardware. The key point in virtual machine performance analysis is to understand how a particular program behaves in a virtualized environment. Focusing on a certain program is well aligned with the opinion of many software developers, because a program is an entity that can be analyzed separatelyand independently, and whose independent behavior is not as non-deterministic as the behavior of the entire system sometimes is. From the perspective of a virtual machine manufacturer, each program is a separate test case revealing problems of virtualization; hence a separate program-centric study is also of importance. The goals of raising the quality of software analysis in virtualizedenvironments to the same level as on real hardware can only be achieved by porting as many advanced performance monitoring methods to a virtual machine as possible. The aforementioned methods include well-known time-based sampling (TBS), popular hardware performance event counting and event-based sampling (EBS), and relatively new statistical call stack reconstruction (statistical call graph), branch...