This paper discusses test techniques used to create an exceptionally reliable high-performance processor using an ultra-dense SRAM cell. We discuss methods to alleviate the read stability fails by enabling the most optimal set of repairs to the highly repairable custom cache arrays without overrunning the total repair capacity of the chip. This paper demonstrates off chip repair calculation strategies to efficiently repair outlier SRAM cells by prioritizing targeted BIST algorithms, test temperature, voltage, and other parameters.

Modern developers rely on container-orchestration frameworks like Kubernetes to deploy and manage hybrid workloads that span the edge and cloud. When network conditions between the edge and cloud change unexpectedly, a workload must adapt its internal behavior. Unfortunately, container-orchestration frameworks do not offer an easy way to express, deploy, and manage adaptation strategies. As a result, fine-tuning or modifying a workload's adaptive behavior can require modifying containers built from large, complex codebases that may be maintained by separate development teams. This paper presents BumbleBee, a lightweight extension for container-orchestration frameworks that separates the concerns of application logic and adaptation logic. BumbleBee provides a simple in-network programming abstraction for making decisions about network data using application semantics. Experiments with a BumbleBee prototype show that edge ML-workloads can adapt to network variability and survive disconnections, edge stream-processing workloads can improve benchmark results between 37.8% and$\boldsymbol{23\mathrm{x}}$, and HLS video-streaming can reduce stalled playback by 77%.

Monitoring extrinsic reliability performance in a given technology is often performed in a passive manner. Semiconductor devices (die) are sampled out of production environment, subjected to a series of electrical and mechanical stresses, and finally tested extensively to determine if stressing produced additional fails. This strategy is prone to missing issues which are localized to a lots or wafers and is not likely to identify potential field reliability problems early enough to remove samples prior to shipment to the customer. We have created a system which continuously scans memory array repairs (correctable errors) across multiple test steps including Burn In and Voltage Screen, applies a pattern classification algorithm, and enables reliability information to be collected for every die. We will show that monitoring extrinsic defects at Wafer Final Test and through Burn In on every die for every product in a technology is crucial to a robust technology reliability strategy. Additionally we demonstrate a low cost high efficiency system, the Active Reliability Monitor, which accomplishes that goal using 22nm POWER8 and zSeries processor data. The system is shown to provide similar quality data to traditional reliability screening methods while also allowing for automated real time reaction to reliability indicators, specific-defect-type monitoring, and robust monitoring across products in the technology. The tool has enabled the identification of new screen procedures used for the containment of several problematic lots and rapid evaluation of the reliability degradation for any suspect reliability vintages.

Public clouds have become a popular platform for building Internet-scale applications. Using virtualization, public cloud services grant customers full control of guest operating systems and applications, while service providers still retain the management of their host infrastructure. Because applications built with public clouds are often highly sensitive to response time, infrastructure builders strive to reduce the latency of their data center's internal network. However, most existing solutions require modification to the software stack controlled by guests. We introduce a new host-centric solution for improving latency in virtualized cloud environments. In this approach, we extend a classic scheduling principle---Shortest Remaining Time First---from the virtualization layer, through the host network stack, to the network switches. Experimental and simulation results show that our solution can reduce median latency of small flows by 40%, with improvements in the tail of almost 90%, while reducing throughput of large flows by less than 3%.

Vehicular environments require continuous awareness of the road ahead. It is critical that mobile applications used in such environments (e.g., GPS route planners and location-based search) do not distract drivers from the primary task of operating the vehicle. Fortunately, a large body of research on vehicular interfaces provides best practices that mobile application developers can follow. However, when we studied the most popular vehicular applications in the Android marketplace, no application followed these guidelines. In fact, vehicular applications were not substantially better at meeting best practice guidelines than non-vehicular applications.

Online social networks have enjoyed significant growth over the past several years. With improvements in mobile and Internet penetration, developing countries are participating in increasing numbers in online communities. This paper provides the first large scale and detailed analysis of social networking usage in developing country contexts. The analysis is based on data from LinkedIn, a professional social network with over 120 million members worldwide. LinkedIn has members from every country in the world, including millions in Africa, Asia, and South America. The goal of this paper is to provide researchers a detailed look at the growth, adoption, and other characteristics of social networking usage in developing countries compared to the developed world. To this end, we discuss several themes that illustrate different dimensions of social networking use, ranging from interconnectedness of members in geographic regions to the impact of local languages on social network participation.

Prefetching is a double-edged sword. It can hide the latency of data transfers over poor and intermittently connected wireless networks, but the costs of prefetching in terms of increased energy and cellular data usage are potentially substantial, particularly for data prefetched incorrectly. Weighing the costs and benefits of prefetching is complex, and consequently most mobile applications employ simple but sub-optimal strategies.

Open and competitive marketplaces, like those enabled by the internet, bring enormous value to consumers while fostering innovation and growth across a wide variety of businesses. Electronic commerce has forever changed how people trade goods and services. Last year alone, US consumers spent more than $150 billion dollars on internet sales [5]. However, developing countries largely have not benefited from these advances, often due to low network penetration, lack of locally relevant markets, and requirements for additional facilities (such as credit cards, shipping arrangements etc.) to take advantage of such marketplaces. Even when these markets are established with local content and poor connectivity in mind, they are often specific to a certain domain or community. This paper introduces Robit, an extensible auction-based market platform for use in challenged network environments. Robit enables developers to incorporate a market layer in their applications and open their service to a wider audience. Robit is built with challenged environments in mind, where communication channels are narrow and potentially expensive. The market structure in Robit is based on a survey of studies in auction theory and economics, and incorporates widely available communication tools. To demonstrate how Robit can be used to add a market layer to digital services, we have modified an open source data fetching application targeted towards challenged environments to use our platform. In addition, we analyze a standalone auction-based marketplace constructed using the Robit infrastructure. We also describe an in-country user study and a small pilot deployment for proof-of-concept.

Mobile devices face a diverse and dynamic set of networking options. Using those options to the fullest requires knowledge of application intent. This paper describes Intentional Networking, a simple but powerful mechanism for handling network diversity. Applications supply a declarative label for network transmissions, and the system matches transmissions to the most appropriate network. The system may also defer and re-order opportunistic transmissions subject to application-supplied mutual exclusion and ordering constraints. We have modified three applications to use Intentional Networking: BlueFS, a distributed file system for pervasive computing, Mozilla's Thunderbird e-mail client, and a vehicular participatory sensing application. We evaluated the performance of these applications using measurements obtained by driving a vehicle through WiFi and cellular 3G network coverage. Compared to an idealized solution that makes optimal use of all aggregated available networks but that lacks knowledge of application intent, Intentional Networking improves the latency of interactive messages from 48% to 13x, while adding no more than 7% throughput overhead.

Developing countries face significant challenges in network access, making even simple network tasks unpleasant. Many standard techniques - caching and predictive prefetching - help somewhat, but provide little or no assistance for personal data that is needed only by a single user. Sulula addresses this problem by leveraging the near-ubiquity of cellular phones able to send and receive simple SMS messages. Rather than visit a kiosk and fetch data on demand - a tiresome process at best - users request a future visit. If capacity exists, the kiosk can schedule secure retrieval of that user's data, saving time and more efficiently utilizing the kiosk's limited connectivity. When the user arrives at a provisioned kiosk, she need only obtain the session key on-demand, and thereafter has instant access. In addition, Sulula allows users to schedule data uploads. Experimental results show significant gains for the end user, saving tens of minutes of time for a typical email/news reading session. We also describe a small, ongoing deployment in-country for proof-of-concept, lessons learned from that experience, and provide a discussion on pricing and marketplace issues that remain to be addressed to make the system viable for developing-world access.

There are many situations in which an additional network interface-or two-can provide benefits to a mobile user. Additional interfaces can support parallelism in network flows, improve handoff times, and provide sideband communication with nearby peers. Unfortunately, such benefits are outweighed by the added costs of an additional physical interface. Instead, virtual interfaces have been proposed as the solution, multiplexing a single physical interface across more than one communication endpoint. However, the switching time of existing implementations is too high for some potential applications, and the benefits of this approach to real applications are not yet clear. This paper directly addresses these two shortcomings. It describes a link-layer implementation of a virtual 802.11 networking layer, called Juggler, that achieves switching times of approximately 3 ms, and less than 400 mus in certain conditions. We demonstrate the performance of this implementation on three application scenarios. By devoting 10 percent of the duty cycle to background tasks, Juggler can provide nearly instantaneous handoff between base stations or support a modest sideband channel with peer nodes, without adversely affecting foreground throughput. Furthermore, when the client issues concurrent network flows, Juggler is able to assign these flows across more than one AP, providing significant speedup when wired-side bandwidth from the AP constrains end-to-end performance.

A quFile is a unifying abstraction that simplifies data management by encapsulating different physical representations of the same logical data. Similar to a quBit (quantum bit), the particular representation of the logical data displayed by a quFile is not determined until the moment it is needed. The representation returned by a quFile is specified by a data-specific policy that can take context into account such as the application requesting the data, the device on which data is accessed, screen size, and battery status. We demonstrate the generality of the quFile abstraction by using it to implement six case studies: resource management, copy-on-write versioning, data redaction, resource-aware directories, application-aware adaptation, and platform-specific encoding. Most quFile policies were expressed using less than one hundred lines of code. Our experimental results show that, with caching and other performance optimizations, quFiles add less than 1% overhead to application-level file system.

Mobile devices cannot rely on a single managed network, but must exploit a wide variety of connectivity options as they travel. We argue that such systems must consider the derivative of connectivity--the changes inherent in movement between separately managed networks, with widely varying capabilities. With predictive knowledge of such changes, devices can more intelligently schedule network usage.

In a peer-to-peer overlay network, hosts cooperate to forward messages. When a message does not reach its final destination, there are two possible explanations. An intermediate overlay host may have dropped the message due to misconfiguration or malice. Alternatively, a bad link in the underlying IP network may have prevented an earnest, properly configured host from forwarding the data. In this paper, we describe how overlay peers can distinguish between the two situations and ascribe blame appropriately. We generate probabilistic notions of blame using distributed network tomography, fuzzy logic, and secure routing primitives. By comparing application-level drop rates with network characteristics inferred from tomography, we can estimate the likelihood that message loss is due to a misbehaving overlay host or a poor link in the underlying IP network. Since faulty nodes can submit inaccurate tomographic data to the collective, we also discuss mechanisms for detecting such misbehavior.

In this paper, we propose a simple yet effective method of identifying traffic conditions on surface streets given location traces collected from on-road vehicles—this requires only GPS location data, plus infrequent low-bandwidth cellular updates. Unlike other systems, which simply display vehicle speeds on the road, our system characterizes unique traffic patterns on each road segment and identifies unusual traffic states on a segment-by-segment basis. We developed and evaluated the system by applying it to two sets of location traces. Evaluation results show that higher than 90 % accuracy in characterization can be achieved after ten or more traversals are collected on a given road segment. We also show that traffic patterns on a road are very consistent over time, provided that the underlying road conditions do not change. This allows us to use a longer history in identifying traffic conditions with higher accuracy.

Abstract Although electronic mail is an increasingly important service, there are few empirical studies of e‐mail traffic. We have observed over 2.85 million messages passing through our departmental servers over the course of seven months, and derived distributions that approximate several important e‐mail parameters including message sizes, message senders and receivers and the burstiness of message deliveries. Our work is unique in that we also analyse message payloads: attachment content types, e‐mail redundancy, and the use of e‐mail as a sharing mechanism. These data can be used in developing e‐mail workloads for mail system engineering or benchmarking. To this end, we provide an improved version of Postmark, a small‐file Internet benchmark, that better approximates mail server characteristics. Copyright © 2007 John Wiley & Sons, Ltd.

This paper presents Virgil, an automatic access point discovery and selection system. Unlike existing systems that select access points based entirely on received signal strength, Virgil scans for all available APs at a location, quickly associates to each, and runs a battery of tests to estimate the quality of each AP's connection to the Internet. Virgil also probes for blocked or redirected ports, to guide AP selection in favor of preserving application services that are currently in use. Results of our evaluation across five neighborhoods in three cities show Virgil finds a usable connection from 22% to 100% more often than selecting based on signal strength alone. By caching AP test results, Virgil both improves performance and success rate. Our overhead is acceptable and is shown to be faster than manually selecting an AP with Windows XP.

In this paper we present a trace-driven framework capable of building realistic mobility models for the simulation studies of mobile systems. With the goal of realism, this framework combines coarse-grained wireless traces, i.e., association data between WiFi users and access points, with an actual map of the space over which the traces were collected. Through a sequence of data processing steps, including filtering the data trace and converting the map to a graph representation, this framework generates a probabilistic mobility model that produces user movement patterns that are representative of real movement. This is done by adopting a set of heuristics that help us infer the paths users take between access points. We describe our experience applying this approach to a college campus, and study a number of properties of the trace data using our framework.

Mobile devices are vulnerable to theft and loss due to their small size and the characteristics of their common usage environment. Since they allow users to work while away from their desk, they are most useful in public locations and while traveling. Unfortunately, this is also where they are most at risk. Existing schemes for securing data either do not protect the device after it is stolen or require bothersome reauthentication. Transient Authentication lifts the burden of authentication from the user by use of a wearable token that constantly attests to the user's presence. When the user departs, the token and device lose contact and the device secures itself. We show how to leverage this authentication framework to secure all the memory and storage locations on a device into which secrets may creep. Our evaluation shows this is done without inconveniencing the user, while imposing a minimal performance overhead

Simulation has become an indispensable tool in the design and evaluation of mobile systems. By using mobility models that describe constituent movement, one can explore large systems, producing repeatable results for comparison between alternatives. In this paper, we show that a large class of mobility models - including all those in which nodal speed and distance or destination are chosen independently - have a transient period in which the average node speed decreases until converging to some long-term average. This speed decay provides an unsound basis for simulation studies that collect results averaged over time, complicating the experimental process. In this paper, we derive a general framework for describing this decay and apply it to a number of cases. Furthermore, this framework allows us to transform a given mobility model into a stationary one by initializing the simulation using the steady-state speed distribution and using the original speed distribution subsequently. This transformation completely eliminates the transient period and the decay in average node speed and, thus, provides sound models for the simulation of mobile systems.

Applications that use parallel TCP streams to increase throughput must multiplex and demultiplex data blocks over a set of TCP streams transmitting on one or more network paths. When applications use the obvious round robin scheduling algorithm for multiplexing data blocks, differences in transmission rate between individual TCP streams can lead to significant data block reordering. This forces the demultiplexing receiver to buffer out-of-order data blocks, consuming memory and potentially causing the receiving application to stall. This paper describes a new adaptive weighted scheduling approach for multiplexing data blocks over a set of parallel TCP streams. Our new scheduling approach, compared with the scheduling approached used by GridFTP, reduces reordering of data blocks between individual TCP streams, maintains the aggregate throughput gains of parallel TCP, consumes less receiver memory for buffering out-of-order packets, and delivers smoother application goodput. We demonstrate the improved characteristics of our new scheduling approach using data transmission experiments over real and emulated wide-area networks.

Unlike the well-administered servers in traditional distributed systems, machines in peer-to-peer networks have widely varying levels of availability. Accurate modeling of node uptime is crucial for predicting per-machine resource burdens and selecting appropriate data replication strategies. In this research project, we improve upon the accuracy of previous peer-to-peer availability models, which are often too conservative to dynamically predict system availability at a fine-grained level. We test our predictors on availability traces from the PlanetLab distributed test bed and the Microsoft corporate network. Each trace has a distinct predictability profile, and we explain these differences by examining the fundamental uptime classes contained in each trace. We also show how availability-guided replica placement reduces the amount of object copying in a distributed data store.

Applications that require good network performance often use parallel TCP streams and TCP modifications to improve the effectiveness of TCP. If the network bottleneck is fully utilized, this approach boosts throughput by unfairly stealing bandwidth from competing TCP streams. Improving the effectiveness of TCP is easy, but improving effectiveness while maintaining fairness is difficult. In this paper, we describe an approach we implemented that uses a long virtual round trip time in combination with parallel TCP streams to improve effectiveness on underutilized networks. Our approach prioritizes fairness at the expense of effectiveness when the network is fully utilized. We compared our approach with standard parallel TCP over a wide-area network, and found that our approach preserves effectiveness and is fairer to competing traffic than standard parallel TCP.

This study examines the random waypoint model widely used in the simulation studies of mobile ad hoc networks. Our findings show that this model fails to provide a steady state in that the average nodal speed consistently decreases over time, and therefore should not be directly used for simulation. We show how unreliable results can be obtained by using this model. In particular, certain ad hoc routing metrics can drop by as much as 40% over the course of a 900-second simulation using the random waypoint model. We give both an intuitive and a formal explanation for this phenomenon. We also propose a simple fix of the problem and discuss a few alternatives. Our modified random waypoint model is able to reach a steady state and simulation results are presented.

Simulation has become an indispensable tool in the construction and evaluation of mobile systems. By using mobility models that describe constituent movement, one can explore large systems, producing repeatable results for comparison between alternatives. Unfortunately, the vast majority of mobility models---including all those in which nodal speed and distance or destination are chosen independently---suffer from decay; average speed decreases until converging to some long-term average. Such decay provides an unsound basis for simulation studies that collect results averaged over time, complicating the experimental process.This paper shows via analysis that such decay is inevitable in a wide variety of mobility models, including the most common in use today. We derive a general framework for describing this decay, and apply it to a number of practical cases. Furthermore, this framework allows us to transform any given mobility model into a stationary one: choose initial speeds from the steady-state distribution, and subsequent speeds from the original. This transformation provides sound models for simulation, eliminating variations in average nodal speed.

How does a machine know who is using it? Current systems authenticate their users infrequently, and assume the user's identity does not change. Such persistent authentication is inappropriate for mobile and ubiquitous systems, where associations between people and devices are fluid and unpredictable. We solve this problem with Transient Authentication, in which a small hardware token continuously authenticates the user's presence over a short-range, wireless link. We present the four principles underlying Transient Authentication, and describe two techniques for securing applications. Applications can be protected transparently by encrypting in-memory state when the user departs and decrypting this state when the user returns. This technique is effective, requiring just under 10 seconds to protect and restore an entire machine, but indiscriminate. Instead, applications can utilize an API for Transient Authentication, protecting only sensitive state. We describe our ports of three applications---PGP, SSH, and Mozilla---to this API. Mozilla, the most complicated application we have ported, suffers less than 4% overhead in page loads in the worst case, and in typical use can be protected in less than 250 milliseconds.

Peer-to-peer storage systems assume that their users consume resources in proportion to their contribution. Unfortunately, users are unlikely to do this without some enforcement mechanism. Prior solutions to this problem require centralized infrastructure, constraints on data placement, or ongoing administrative costs. All of these run counter to the design philosophy of peer-to-peer systems.Samsara enforces fairness in peer-to-peer storage systems without requiring trusted third parties, symmetric storage relationships, monetary payment, or certified identities. Each peer that requests storage of another must agree to hold a claim in return---a placeholder that accounts for available space. After an exchange, each partner checks the other to ensure faithfulness. Samsara punishes unresponsive nodes probabilistically. Because objects are replicated, nodes with transient failures are unlikely to suffer data loss, unlike those that are dishonest or chronically unavailable. Claim storage overhead can be reduced when necessary by forwarding among chains of nodes, and eliminated when cycles are created. Forwarding chains increase the risk of exposure to failure, but such risk is modest under reasonable assumptions of utilization and simultaneous, persistent failure.

Laptops are vulnerable to theft, greatly increasing the likelihood of exposing sensitive files. Unfortunately, storing data in a cryptographic file system does not fully address this problem. Such systems ask the user to imbue them with long-term authority for decryption, but that authority can be used by anyone who physically possesses the machine. Forcing the user to frequently reestablish his identity is intrusive, encouraging him to disable encryption.Our solution to this problem is Zero-Interaction Authentication, or ZIA. In ZIA, a user wears a small authentication token that communicates with a laptop over a short-range, wireless link. Whenever the laptop needs decryption authority, it acquires it from the token; authority is retained only as long as necessary. With careful key management, ZIA imposes an overhead of only 9.3% for representative workloads. The largest file cache on our hardware can be re-encrypted within five seconds of the user's departure, and restored in just over six seconds after detecting the user's return. This secures the machine before an attacker can gain physical access, but recovers full performance before a returning user resumes work.

The use of parallel TCP connections to increase throughput for bulk transfers is common practice within the high performance computing community. However, the effectiveness, fairness, and efficiency of data transfers across parallel connections is unclear. This paper considers the impact of systemic non-congestion related packet loss on the effectiveness, fairness, and efficiency of parallel TCP transmissions. The results indicate that parallel connections are effective at increasing aggregate throughput, and increase the overall efficiency of the network bottleneck. In the presence of congestion related losses, parallel flows steal bandwidth from other single stream flows. A simple modification is presented that reduces the fairness problems when congestion is present, but retains effectiveness and efficiency.

This paper argues that the operating system and applications currently running on a real machine should relocate into a virtual machine. This structure enables services to be added below the operating system and to do so without trusting or modifying the operating system or applications. To demonstrate the usefulness of this structure, we describe three services that take advantage of it: secure logging, intrusion prevention and detection, and environment migration.

Mobile users can increasingly depend on high speed connectivity. Despite this, using distributed file services across the wide area is painful. Fast approaches sacrifice one or more of safety, visibility, and consistency in the name of performance. Instead, we propose fluid replication, the ability to create replicas where and when needed. These replicas, called WayStations, maintain consistency with home servers through periodic reconciliations. Two techniques make reconciliation fast; this is crucial to the success of fluid replication. First, we defer propagation of updates, and only invalidate files during a reconciliation. Second, rather than depend on operation logs, we provide the subtrees in which all updates have occurred. These subtrees, named by their least common ancestors, or LCAs, can be constructed incrementally, and reduce the burden of checking serializability during a reconciliation. While these techniques provide better performance, they are not without risk. Bulk invalidation can lead to false sharing, optimistic updates are subject to conflict, and deferred updates may cause performance problems if they are needed elsewhere. To address these concerns, we performed a trace-based evaluation of our algorithms.

Mobile systems must adapt their behavior to changing network conditions. To do this, they must accurately estimate available network capacity. Producing quality estimates is challenging because network observations are noisy, particularly in mobile, ad hoc networks. Current systems depend on simple, exponentially-weighted moving average (EWMA) filters. These filters are either able to detect true changes quickly or to mask observed noise and transients, but cannot do both. In this paper, we present four filters designed to react quickly to persistent changes while tolerating transient noise. Such filters are agile when possible, but stable when necessary, adapting their behavior to prevailing conditions. These filters are evaluated in a variety of networking situations, including persistent and transient change, congestion, and topology changes. We find that one filter, based on techniques from statistical process control provides performance superior to the other three. Compared to two EWMA filters, one agile and the other stable, it is able to offer the agility of the former in four of five scenarios and the stability of the latter in three of four scenarios.

Subjecting a mobile computing system to wireless network conditions that are realistic yet reproducible is a challenging problem. In this paper, we describe a technique called trace modulation that re-creates the observed end-to-end characteristics of a real wireless network in a controlled and repeatable manner. Trace modulation is transparent to applications and accounts for all network traffic sent or received by the system under test. We present results that show that it is indeed capable of reproducing wireless network performance faithfully.

In this paper we show that application-aware adaptation, a collaborative partnership between the operating system and applications, offers the most general and effective approach to mobile information access.We describe the design of Odyssey, a prototype implementing this approach, and show how it supports concurrent execution of diverse mobile applications.We identify agility as a key attribute of adaptive systems, and describe how to quantify and measure it.We present the results of our evaluation of Odyssey, indicating performance improvements up to a factor of 5 on a benchmark of three applications concurrently using remote services over a network with highly variable bandwidth.This research was supported by the

In this paper we present results from a six-month empirical study of the high availability aspects of the Coda File System. We report on the service failures experienced by Coda clients, and show that such failures are masked successfully. We also explore the effectiveness and resource costs of key aspects of server replication and disconnected operation, the two high availability mechanisms of Coda. Wherever possible, we compare our measurements to simulation-based predictions from earlier papers and to anecdotal evidence from users. Finally, we explore how users take advantage of the support provided by Coda for mobile computing.