Mahadev Satyanarayanan
Mahadev Satyanarayanan is an experimental computer scientist, a founding pioneer in edge computing, mobile computing, pervasive computing, and Internet of Things. He is the Carnegie Group Professor of Computer Science at Carnegie Mellon University. Satya obtained his Bachelor's and Master's degree from Indian Institute of Technology, Madras in 1975 and 1977, and his Ph.D. in Computer Science from CMU in 1983. He is an ACM and IEEE fellow. His work on the Andrew File System was recognized with the prestigious ACM Software System Award in 2016 and the ACM SIGOPS Hall of Fame Award in 2008 for its excellent engineering and long-lasting impact. His work on disconnected operation in Coda File System received the ACM SIGOPS Hall of Fame Award in 2015 and the inaugural ACM SIGMOBILE Test-of-Time Award in 2016. He served as the founding Program Chair of the and the , the founding Editor-in-Chief of , and the founding Area Editor for the . In addition, he was the founding director of Intel Research Pittsburgh and an advisor to the company Maginatics, which was acquired by EMC in 2014.
In his research career, Satya has focused on the performance, scalability, availability and trust challenges in computing systems that reach from the cloud to the mobile edge of the Internet. He has pioneered many advances in edge computing, distributed systems, mobile computing, pervasive computing, and the Internet of Things. Below is a chronological summary of his primary research contributions.
Andrew File System
Satya was a principal architect and implementer of the Andrew File System, the technical forerunner of modern cloud-based storage systems. AFS has been continuously deployed at CMU since 1986, at a scale of many thousands of users. From its conception in 1983 as the unifying campus-wide IT infrastructure for CMU, AFS evolved through versions AFS-1, AFS-2 and AFS-3. In mid-1989, AFS-3 was commercialized by Transarc Corporation and its evolution continued outside CMU. Transarc was acquired by IBM, and AFS became an IBM product for a number of years. In 2000, IBM released the code to the open source community as OpenAFS. Since its release as OpenAFS, the system has continued to be used in many enterprises all over the world. In the academic and research lab community, OpenAFS is in use at more than 30 sites in the United States and dozens of sites in Europe, New Zealand, and South Korea. Many global companies have used OpenAFS including Morgan Stanley, Goldman Sachs, Qualcomm, IBM, United Airlines, Pfizer, Hitachi, InfoPrint, and Pictage.Over a 30-year period, AFS has been a seminal influence on academic research and commercial practice in distributed data storage systems for unstructured data. Its approach to native file system emulation, scalable file caching, access-control based security, and scalable system administration have proved to be of enduring value in enterprise-scale information sharing. The design principles that were initially discovered and validated in the creation and evolution of AFS have influenced virtually all modern commercial distributed file systems, including Microsoft DFS, Google File System, Lustre File System, Ceph, and NetApp ONTAP. In addition, AFS inspired the creation of DropBox whose founders used AFS as part of Project Athena at MIT. It also inspired the creation of Maginatics, a startup company advised by Satya that provides cloud-sourced network-attached storage for distributed environments. The NFS v4 network file system protocol standard has been extensively informed by the lessons of AFS. In 2016, AFS was honored with the prestigious . Earlier, ACM recognized the significance of AFS by inducting a key paper on it to the ACM SIGOPS Hall of Fame. The AFS papers in 1985 and 1987 also received Outstanding Paper awards at the ACM Symposium on Operating System Principles.
Coda File System
In 1987, Satya began work on the Coda File System to address a fundamental shortcoming of AFS-like systems. Extensive first-hand experience with the AFS deployment at CMU showed that users are severely impacted by server and network failures. This vulnerability is not just hypothetical, but indeed a fact of life in real-world deployments. Once users become critically dependent on files cached from servers, a server or network failure renders these files inaccessible and leaves clients crippled for the duration of the failure. In a large enough system, unplanned outages of servers and network segments are practically impossible to avoid. Today's enthusiastic embrace of cloud computing rekindles many of these concerns because of increased dependence on centralized resources, The goal of the Coda project was to preserve the many strengths of AFS, while reducing its vulnerability to failures. Over 30+ years, research on Coda has proved to be highly fruitful in creating new insights and mechanisms for failure-resilient, scalable and secure read-write access to shared information by mobile and static users over wireless and wired networks. Coda was the first system to show how server replication could be combined with client caching to achieve good performance and high availability. Coda invented the concept of "disconnected operation", in which cached state on clients is used to mask network and server failures. Coda also demonstrated bandwidth-adaptive weakly-connected operation over networks with low bandwidth, high latency or frequent failures. Coda's use of optimistic replication, trading consistency for availability, was controversial when introduced. Today, it is a standard practice in all data storage systems for mobile environments. Coda also pioneered the concept of translucent caching, which balances the full transparency of classic caching with the user visibility needed to achieve a good user experience on bandwidth-challenged networks. The Coda concepts of hoarding, reintegration and application-specific conflict resolution are found in the cloud sync capabilities of virtually all mobile devices today. Key ideas from Coda were incorporated by Microsoft into the component of Windows 2000 and the . Papers relating to Coda received Outstanding Paper awards at the and ACM Symposium on Operating System Principles. In 1999, Coda received the . A 2002 narrative retrospective, "" traces its evolution and the lessons learned from it. Later, Coda's long-lasting impact was recognized with the ACM SIGOPS Hall of Fame Award in 2015 and the inaugural ACM SIGMOBILE Test-of-Time Award in 2016.Odyssey: Application-aware Adaptation for Mobile Applications
In the mid-1990s, Satya initiated the Odyssey project to explore how operating systems should be extended to support future mobile applications. While Coda supported mobility in an application-transparent manner, Odyssey explored the space of application-aware approaches to mobility. Wireless network bandwidth and energy were two of the key resource challenges faced by mobile applications. Odyssey invented the concept of application-aware adaptation and showed how the system call interface to the Unix operating system could be extended to support this new class of mobile applications such as video delivery and speech recognition. Odyssey envisioned a collaborative partnership between the operating system and individual applications. In this partnership, the operating system monitors, controls and allocates scarce resources such as wireless network bandwidth and energy, while the individual applications negotiate with the operating system on their resource requirements and modify application behavior to offer the best user experience achievable under current resource conditions. The and Odyssey papers on application-aware adaptation and energy-aware adaptation in the ACM Symposium on Operating System Principles have proved to be highly influential. The concepts of multi-fidelity algorithms and predictive resource management that emerged from this work have also proved to be influential.Aura: Cloud Offload for IoT
In the late 1990s, Satya initiated the Aura Project in collaboration with CMU faculty colleagues David Garlan, Raj Reddy, Peter Steenkiste, Dan Siewiorek and Asim Smailagic. The challenge addressed by this effort was to reduce human distraction in mobile and pervasive computing environments, recognizing that human attention does not benefit from Moore's Law, while computing resources do. This leads directly to the notion of invisible computing, which parallels Mark Weiser's characterization of an ideal technology as one that disappears. The Aura vision proved to be an excellent driver of research in mobile and pervasive computing in areas such as cyber foraging, location-aware computing, energy-awareness, and task-level adaptation. In particular, the 1997 paper "" pioneered "cloud offload," in which mobile devices transmit processed sensor data to a cloud service for further analysis over a wireless network. A modern incarnation of this idea is speech recognition using Siri. Specifically, a user's speech is captured by a microphone, pre-processed, and then sent to a cloud service that converts speech to text. Satya continues to do IoT-related research. He retrospectively described the evolutionary path from his early work to today's cloud-based mobile and IoT systems in "".Reflecting on the Aura vision and IoT implementation experience to date, Satya wrote an invited paper in 2001 entitled "" This has proved to be his most widely cited work according to Google Scholar, and continues to receive well over 100 citations each year. The concepts discussed in this paper have directly inspired today's popular vision of an "Internet of Things." In 2018, this visionary paper was recognized by the ACM SIGMOBILE Test-of-Time Award.
Internet Suspend/Resume (ISR): Virtual Desktop
Building on Intel's newly-available VT virtual machine technology in 2001, ISR represents an AFS-like capability for cloud-sourced VMs. Instead of just delivering files, ISR enables entire computing environments to be delivered from the cloud with perfect fidelity through on-demand caching to the edges of the Internet. introducing the ISR concept was the first to articulate the concept of wide-area hands-free mobile computing with a "zero-pound laptop." The ISR concept has proved to be highly influential in the mobile computing research community, spawning related research efforts in industry and academia. A series of implementations and associated deployments of ISR at CMU have investigated the implementation trade-offs in this space and demonstrated the real-world viability of this technology. The ISR project inspired commercial software such as Citrix XenDesktop and Microsoft Remote Desktop Services, commonly known as Virtual Desktop Infrastructure. The VDI industry has since became a billion-dollar industry.Olive: Execution Fidelity for Software Archiving
The work on ISR inspired the Olive project, a collaboration between the computer science and digital library communities. One of the major challenges of digital archiving is the ability to preserve and accurately reproduce executable content across time periods of many decades. This problem also has analogs in industry. For example, a NASA space probe to the edge of the solar system may take 30 years to reach its destination; software maintenance over such an extended period requires precise re-creation of the probe's onboard software environment. By encapsulating the entire software environment in a VM, Olive preserves and dynamically reproduces the precise execution behavior of software. The Olive prototype demonstrated reliable archiving of software dating back to the early 1980s. The concept of , introduced by Olive, has proved to be highly influential in digital archiving.Diamond: Unindexed Search for High-dimensional Data
The explored interactive search of complex data such as photographs, video, and medical images that have not been tagged or indexed a priori. For such unstructured and high-dimensional data, the classical approach of full-text indexing is not viable: in contrast to text, which is human-authored and one-dimensional, raw image data requires a feature extraction step prior to indexing. Unfortunately, the features to extract for a given search are not known a priori. Only through interactive trial and error, with partial results to guide his progress, can a user converge on the best choice of features for a specific search. To support this search workflow, the platform provided a storage architecture for discard-based search that pipelines user control, feature extraction, and per-object indexing computation and result caching. As documented in , the I/O workloads generated by Diamond searches differ significantly from well-understood indexing workloads such as Hadoop, with important implications for storage subsystems. The unique search capabilities of Diamond attracted significant interest in the medical and pharmaceutical research communities. Researchers in these communities collaborated in creating Diamond-based applications for domains such as radiology, pathology and dermatology, drug discovery, and craniofacial genetics. The work on Diamond and associated software spurred extensive collaboration between Satya's research group at CMU and Health Sciences at the University of Pittsburgh. The collaboration with pathologists led to the design and implementation of a vendor‐neutral open source library for digital pathology. OpenSlide is in use today by many academic and industrial organizations worldwide, including many research sites in the United States that are funded by the National Institutes of Health and companies such as .Elijah: Edge Computing
Satya pioneered edge computing with the 2009 publication of the paper "," and his ensuing research efforts in . This paper is now widely recognized as the founding manifesto of edge computing, and has proved to be highly influential in shaping thoughts and actions. It was written in close collaboration with Victor Bahl from Microsoft, Roy Want from Intel, Ramon Caceres from AT&T, and Nigel Davies from Lancaster University. This paper introduced the concept of cloudlets, which are small data-centers located at the network edge. As a new computing tier between mobile devices and the cloud, they have powerful computational resources and excellent connectivity to mobile devices, typically just one wireless hop away. Their low latency and high bandwidth to mobile users and sensors make them ideal locations for offloading computation. A detailed account of the origin of the paper and the cloudlet concept is described in the 2014 retrospective, "."Edge computing has now become one of the hottest topics in industry and academia. It is particularly relevant to mobile and IoT use cases in which a significant amount of live sensor data needs to be intensively processed in real-time. Many applications in domains such as VR/AR, factory automation, and autonomous vehicles exhibit such workflow. For example, high-quality commercial VR headsets, such as Oculus Rift and HTC Vive, require tethering to a GPU-equipped desktop. Such tethering negatively impacts user experience. On the other hand, untethered devices sacrifice the quality of the virtual environment. Fundamentally, these latency-sensitive, resource-hungry, and bandwidth-intensive applications cannot run on mobile devices alone due to insufficient compute power, nor can they run in the cloud due to long network latency. Only edge computing can break this deadlock.
The concepts of VM synthesis and VM handoff were conceived and demonstrated in Elijah, leading to the reference implementation of cloudlet software infrastructure. The is a collection of companies working closely with CMU to build an open ecosystem for edge computing.