Neal H. Walfield

There is a blog.

Papers

Smart phones should not just accompany their owners: they should provide them with the data they want whenever and wherever they are. This does not mean that the user should always be able to fetch data on demand: wireless communication requires a significant amount of energy; cellular bandwidth is often limited; and, coverage is not ubiquitous. Instead, scheduling data stream updates, e.g., podcast downloads and photo uploads, should incorporate predictions of where the user will be, when and where data will be needed, and when transfer conditions are good (e.g., WiFi is available). The challenges we have encountered while designing such a framework include scheduling transmissions while respecting multiple optimization goals, application integration, modeling user and data stream behavior, and managing prefetched data. To better understand user behavior and evaluate our framework, we are gathering traces of smart phone use. Initial results show potential energy savings of over 70%.

We present a storage management framework for Web 2.0 services that places users back in control of their data. Current Web services complicate data management due to data lock-in and lack usable protection mechanisms, which makes cross-service sharing risky. Our framework allows multiple Web services shared access to a single copy of data that resides on a personal storage repository, which the user acquires from a cloud storage provider. Access control is based on hierarchically, filtered views, which simplify cross-cutting policies, and enable least privilege management. We also integrate a powerbox, which allows applications to request additional authority at run time thereby enabling applications running under a least privilege regime to provide useful open and save as dialogs.

Many programs could improve their performance by adapting their memory use according to availability. If memory is available, a web browser or DNS server could use a larger cache; if there is memory pressure, they could use a smaller one. Memory adaptations are also becoming increasingly important for scalability: server consolidation is being done more aggressively and end-users want to run their application on a wider-range of hardware configurations. Today, memory-based adaptations are fragile: general-purpose operating systems do not indicate how much memory a process should or could use.

Enabling efficient adaptations requires rethinking how memory is allocated among competing programs, and adding a feedback mechanism that allows applications to make informed adaptations. In this paper, we present the design and implementation of a minimum-funding revocation scheduler for memory. We describe a novel algorithm to compute the amount of memory available to each resource principal based on its scheduling parameters and the current configuration, explain how to communicate this information to the principals, and show how they can exploit it. We also present a new mechanism to account shared memory based on access frequency. We demonstrate the effectiveness of the techniques by showing that multiple applications changed to exploit this information use the full memory available to them, and smoothly vary their demand as availability changes. This also results in significant increases in throughput relative to the conservative management techniques currently used.

Many programs could improve their performance by adapting their memory use. To ensure that an adaptation increases utility, a program needs to know not only how much memory is available but how much it should use at any given time. This is difficult on commodity operating systems, which provide little information to inform these decisions. As evidence of the importance of adaptations to program performance, many programs currently adapt using ad hoc heuristics to control their behavior.

Supporting adaptive applications has become pressing: the range of hardware that applications are expected to run on---from smart phones and netbooks to high-end desktops and servers---is increasing as is the dynamic nature of workloads stemming from server consolidation. The practical result is that the ad hoc heuristics are less effective as assumptions about the environment are less reliable, and as such memory is more frequently under- or over- utilized. Failing to adapt limits the degree of possible consolidation. We contend that in order for programs to make the best of available resources, research needs to be conducted into how the operating system can better support aggressive adaptations.

General-purpose operating systems not only fail to provide adaptive applications the information they need to intelligently adapt, but also schedule resources in such a way that were applications to aggressively adapt, resources would be inappropriately scheduled. The problem is that these systems use demand as the primary indicator of utility, which is a poor indicator of utility for adaptive applications.

We present a resource management framework appropriate for traditional as well as adaptive applications. The primary difference from current schedulers is the use of stakeholder preferences in addition to demand. We also show how to revoke memory, compute the amount of memory available to each principal, and account shared memory. Finally, we introduce a prototype system, Viengoos, and present some benchmarks that demonstrate that it can efficiently support multiple aggressively adaptive applications simultaneously.

Current operating systems provide inadequate mechanisms to protect user data. The main problem is that all of a user's programs run in the same trust domain. A better model is one which is consistent with the principle of least authority (POLA). An object-capability system may be able to better achieve this: capabilities bundle authorization and designation thereby easing delegation and the dynamic creation and management of fine-grained trust domains.

Despite this, object-capability designs are rejected due to a perceived excessive overhead resulting from the degree of decomposition and the corresponding rise in the amount of inter-process communication (IPC). Although the work on L4 has demonstrated that IPC can be made extremely fast, historically, L4 lacks mechanisms to efficiently delegate fine-grained authority.

In this paper, we present a capability transfer mechanism that exploits the memory management unit (MMU) present in all modern commodity hardware by using it to build a content addressable memory (CAM) to expedite capability resolution. For the common case of an IPC carrying a single capability, we observe a 2% increase in message transfer time compared to a less flexible but more commonly used IPC implementation based on capability registers. Relative to the time taken to transfer a similarly sized message containing just data on L4Ka::Pistachio, we observe a 16% increase.

The GNU Hurd's design was motivated by a desire to rectify a number of observed shortcomings in Unix. Foremost among these is that many policies that limit users exist simply as remnants of the design of the system's mechanisms and their implementation. To increase extensibility and integration, the Hurd adopts an object-based architecture and defines interfaces, in particular those for the composition of and access to name spaces, that are virtualizable.

This paper is first a presentation of the Hurd's design goals and a characterization of its architecture primarily as it represents a departure from Unix's. We then critique the architecture and assess it in terms of the user environment of today focusing on security. Then follows an evaluation of Mach, the microkernel on which the Hurd is built, emphasizing the design constraints which Mach imposes as well as a number of deficiencies its design presents for multi-server like systems. Finally, we reflect on the properties such a system appears to require.

Commodity operating systems fail to meet the security, resource management and integration expectations of users. We propose a unified solution based on a capability framework as it supports fine grained objects, straightforward access propagation and virtualizable interfaces and explore how to improve resource use via access decomposition and policy refinement with minimum interposition. We argue that only a small static number of scheduling policies are needed in practice and advocate hierarchical policy specification and central realization.

  • Coyotos Microkernel Specification with Jonathan S. Shapiro, Ph.D., Eric Northup, M. Scott Doerrie, Swaroop Sridhar and Marcus Brinkmann. Version 0.3+ published online March 20, 2006.

Coyotos is a security microkernel. It is a microkernel in the sense that it is a minimal protected platform on which a complete operating system can be constructed. It is a security microkernel in the sense that it is a minimal protected platform on which higher level security policies can be constructed.

Through the use of a multiserver capability-based architecture, the GNU Hurd has attempted to increase security and flexibility relative to traditional Unix-like operating systems. This shift away from a monolithic design requires a reevaluation of conventional operating system praxis to determine its degree of continued applicability. Resource scheduling appears particularly defunct in this regard: to make smarter scheduling decisions, monolithic systems cross component boundaries to gain insight into application behavior. This introspection is incompatible with a multiserver architecture and its elimination, as observed in Mach, the current microkernel used by the GNU Hurd, results in noticeable performance degradation. To this end, I propose that rather than have the operating system provide virtualized resources, i.e.~schedule the contents of resources on behalf of applications, it offer near raw access to the principals which they must multiplex as required thereby relieving e.g.~the memory manager of paging decisions. The resource managers must still partition the physical resources among the competing principals. For this, I suggest a market based solution in which principals have a periodically renewed credit allowance and lease the required resources. This approach also suits adaptive and soft real-time applications.

Intellectual property presupposes the individuality of the author by asserting that the author is unique and is able to transcend the realm of the mundane to discover a novel idea. Intellectual property denies T.S. Elliot's assertion in Tradition and the Individual Talent that the ``poet's mind is in fact a receptacle for seizing and storing up numberless feelings, phrases, images, which remain there until all the particles which can unite to form a new compound are present together.'' Intellectual property is a logical impossibility for its fundamental tenant is that thoughts and ideas, once shared, can be owned and controlled by an individual.

Contact

You can email me at: neal@walfield.org .

If you send mail to spamtrap@walfield.org, it will not reach me and be automatically classified as spam.