Virtualization Technologies & Resource Management
Resource management is a key component of all modern operating systems and hypervisors. In the context of virtualization, the ability to manage resources efficiently is critical for giving each virtual machine (VM) the illusion of being a dedicated physical machine that is fully protected and isolated from other virtual machines. Virtualization platforms should support flexible overcommitment of processor, memory, and other resources in order to reap the benefits of statistical multiplexing, while still providing quality-of-service guarantees to VMs of varying importance. New challenges and opportunities arise as virtualization systems and the hardware they manage continue to evolve. In June 2006, VMware held its second seminar to discuss resource management relevant to the virtualization space. Abstracts and links to the selected presentations are below.
Selected Presentation Abstracts
Resource Management for Virtualized Systems – Carl Waldspurger, VMware R&D
Virtualization platforms must manage hardware resources efficiently, and provide flexible quality-of-service controls for virtual machines of varying importance. This presentation will describe key challenges and opportunities related to managing diverse resources for virtualized systems. We will briefly examine several resource management topics, including resource controls, processor scheduling, memory management, NUMA scheduling, and distributed load balancing. For each topic, we will examine useful features and current challenges, describe aspects of VMware's existing solutions, and highlight opportunities for future research.
Hijack: Taking Control of COTS Systems to Enforce Predictable Service Guarantees – Richard West and Gabriel Parmer, Boston University
In order to eliminate the costs of proprietary systems and special purpose hardware, many real-time and embedded computing platforms are being built on commodity operating systems and generic hardware. Unfortunately, many such systems are ill-suited to the low-latency and predictable timing requirements of real-time applications. This presentation, therefore, focuses on application-specific service technologies for low-cost commodity operating systems and hardware, so that real-time service guarantees can be met. We describe contrasting methods to deploy first-class services on commodity systems that are dispatched with low latency and execute asynchronously according to bounds on CPU, memory and I/O device usage.
The focus of this presentation is on user-level techniques to isolate application-specific services that can be invoked in a safe, predictable and efficient manner. We provide some background to our Hijack software architecture by first describing our earlier work on system extensibility using "SafeX" and "user-level sandboxing". Finally, we describe how our Hijack system can be used to effectively take control of COTS systems for predictable service execution and resource management without modifications to the underlying kernel.
Friendly Virtual Machines - Leveraging a Feedback-Control Model for Application Adaptation – Azer Bestavros and Yuting Zhang, Boston University
With the increased use of "Virtual Machines" (VMs) as vehicles that isolate applications running on the same host, it is necessary to devise techniques that enable multiple VMs to share underlying resources both fairly and efficiently. To that end, one common approach is to deploy complex resource management techniques in the hosting infrastructure. Alternately, in this talk, we advocate the use of self adaptation in the VMs themselves based on feedback about resource usage and availability. Consequently, we define a "Friendly" VM (FVM) to be a virtual machine that adjusts its demand for system resources, so that these resources end-up being both efficiently and fairly allocated to competing FVMs.
Such properties are ensured using one of many provably convergent control rules, such as Additive-Increase/Multiplicative-Decrease (AIMD). By adopting this distributed application-based approach to resource management, it is not necessary to make assumptions neither about the underlying resources nor about the requirements of FVMs competing for these resources. To demonstrate the elegance and simplicity of our approach, we present a prototype implementation of our FVM framework in User-Mode Linux (UML)-an implementation that consists of less than 500 lines of code changes to UML. We present an analytic, control-theoretic model of FVM adaptation, which establishes convergence and fairness properties. These properties are also backed up with experimental results using our prototype FVM implementation.
This work was done in collaboration with Mina Guirguis, Ibrahim Matta, and Rich West.
Speaker Biographies
Azer Bestavros obtained his SM in 1988 and his Ph.D. in 1992, both in Computer Science from Harvard University . He is currently Professor and Chairman of Computer Science at Boston University . His research interests are in the general areas of networking and real-time systems. Some of his seminal works include his generalization of classical rate-monotonic analysis to accommodate probabilistic guarantees, his pioneering of the push model for Internet content distribution adopted years later by CDNs, and his characterization of Web traffic self-similarity and reference locality. With over 2,000 citations to his publications, CiteSeer ranks him in the top 250 of its 10,000 most-cited CS authors at all times. His research has been funded by government and industry grants totaling over $15M. He received distinguished service awards from both the ACM and the IEEE, and is a distinguished speaker of the IEEE.
Gabriel Parmer is a Ph.D. student at Boston University , working on topics related to operating systems, real-time systems and resource management. He currently holds a Bachelor of Arts degree from Boston University , which he received in 2003. His current research focus is on the topic of structuring systems to support safe, predictable and efficient application-specific services. Notable publications on this topic include a best paper at IEEE RTAS 2006 and also Cluster Computing 2004.
Carl Waldspurger is a Principal Engineer at VMware, responsible for core resource management and virtualization technologies. Prior to joining VMware, he was a researcher at the DEC/Compaq Systems Research Center. He completed his Ph.D. in computer science at MIT in 1995, receiving the ACM doctoral dissertation award. He has published numerous papers in the areas of resource management, virtualization, and operating systems, and is an inventor with over twenty issued patents.
Richard West received the MEng degree in microelectronics and software engineering in 1991 from the University of Newcastle-upon-Tyne, England. He later received both the MS (1998) and Ph.D. (2000) degrees in computer science from the Georgia Institute of Technology. He is currently an assistant professor in the Computer Science Department at Boston University , where his research interests include operating systems, real-time systems, distributed computing and QoS management.
Yuting Zhang is a Ph.D. candidate in Computer Science Department at Boston University . She got her Bachelor's and Master's degrees in Computer Science from the University of Science and Technology Beijing.
She joined Boston University in 2001 and is expected to graduate in August 2006. During the summer of 2005 and Spring 2006, she worked at VMware as an intern on the timer sponge project under the supervision of Carl Waldspurger. Her research interests are in the general areas of operating systems and networking, including scheduling in multimedia and embedded real-time systems, resource management in virtual execution environments, and network security.
