Search

The current design trend in large scale machine learning is to use distributed clusters of CPUs and GPUs with MapReduce-style programming. Some have been led to believe that this type of horizontal scaling can reduce or even eliminate the need for traditional algorithm development, careful parallelization, and performance engineering. This paper is a case study showing the contrary: that the benefits of algorithms, parallelization, and performance engineering, can sometimes be so vast that it is possible to solve "cluster-scale" problems on a single commodity multicore machine. Connectomics is an emerging area of neurobiology that uses cutting edge machine learning and image processing to extract brain connectivity graphs from electron microscopy images. It has long been assumed that the processing of connectomics data will require mass storage, farms of CPU/GPUs, and will take months (if not years) of processing time. We present a high-throughput connectomics-on-demand system that runs on a multicore machine with less than 100 cores and extracts connectomes at the terabyte per hour pace of modern electron microscopes., National Science Foundation (U.S.) (grant IIS-1447786), National Science Foundation (U.S.) (grant CCF1563880), United States. Intelligence Advanced Research Projects Activity (grant 138076-5093555)

Git is a widely used version control system that is powerful but complicated. Its complexity may not be an inevitable consequence of its power but rather evidence of flaws in its design. To explore this hypothesis, we analyzed the design of Git using a theory that identifies concepts, purposes, and misfits. Some well-known difficulties with Git are described, and explained as misfits in which underlying concepts fail to meet their intended purpose. Based on this analysis, we designed a reworking of Git (called Gitless) that attempts to remedy these flaws. To correlate misfits with issues reported by users, we conducted a study of Stack Overflow questions. And to determine whether users experienced fewer complications using Gitless in place of Git, we conducted a small user study. Results suggest our approach can be profitable in identifying, analyzing, and fixing design problems., SUTD-MIT International Design Centre (IDC)

We present a method for synthesizing recursive functions that provably satisfy a given specification in the form of a polymorphic refinement type. We observe that such specifications are particularly suitable for program synthesis for two reasons. First, they offer a unique combination of expressive power and decidability, which enables automatic verification—and hence synthesis—of nontrivial programs. Second, a type-based specification for a program can often be effectively decomposed into independent specifications for its components, causing the synthesizer to consider fewer component combinations and leading to a combinatorial reduction in the size of the search space. At the core of our synthesis procedure is a newalgorithm for refinement type checking, which supports specification decomposition. We have evaluated our prototype implementation on a large set of synthesis problems and found that it exceeds the state of the art in terms of both scalability and usability. The tool was able to synthesize more complex programs than those reported in prior work (several sorting algorithms and operations on balanced search trees), as well as most of the benchmarks tackled by existing synthesizers, often starting from a more concise and intuitive user input., National Science Foundation (U.S.) (Grant CCF-1438969), National Science Foundation (U.S.) (Grant CCF-1139056), United States. Defense Advanced Research Projects Agency (Grant FA8750-14-2-0242)

High-level scripting languages have become tremendously popular for development of dynamic Web applications. Many programmers appreciate the productivity benefits of automatic storage management, freedom from verbose type annotations, and so on. While it is often possible to improve performance substantially by rewriting an application in C or a similar language, very few programmers bother to do so, because of the consequences for human development effort. This paper describes a compiler that makes it possible to have most of the best of both worlds, coding Web applications in a high-level language but compiling to native code with performance comparable to handwritten C code. The source language is Ur/Web, a domain-specific, purely functional, statically typed language for the Web. Through a coordinated suite of relatively straightforward program analyses and algebraic optimizations, we transform Ur/Web programs into almost-idiomatic C code, with no garbage collection, little unnecessary memory allocation for intermediate values, etc. Our compiler is in production use for commercial Web sites supporting thousands of users, and microbenchmarks demonstrate very competitive performance versus mainstream tools.

A daunting challenge faced by program performance autotuning is input sensitivity, where the best autotuned configuration may vary with different input sets. This paper presents a novel two-level input learning algorithm to tackle the challenge for an important class of autotuning problems, algorithmic autotuning. The new approach uses a two-level input clustering method to automatically refine input grouping, feature selection, and classifier construction. Its design solves a series of open issues that are particularly essential to algorithmic autotuning, including the enormous optimization space, complex influence by deep input features, high cost in feature extraction, and variable accuracy of algorithmic choices. Experimental results show that the new solution yields up to a 3x speedup over using a single configuration for all inputs, and a 34x speedup over a traditional one-level method for addressing input sensitivity in program optimizations., United States. Dept. of Energy (Award DE-SC0005288), United States. Dept. of Energy (Award DE-SC0008923), United States. Dept. of Energy (Early Career Award), National Science Foundation (U.S.) (Grant 1464216), National Science Foundation (U.S.) (Grant 1320796), National Science Foundation (U.S.) (CAREER Award)

Many verifications of realistic software systems are monolithic, in the sense that they define single global invariants over complete system state. More modular proof techniques promise to support reuse of component proofs and even reduce the effort required to verify one concrete system, just as modularity simplifies standard software development. This paper reports on one case study applying modular proof techniques in the Coq proof assistant. To our knowledge, it is the first modular verification certifying a system that combines infrastructure with an application of interest to end users. We assume a nonblocking API for managing TCP networking streams, and on top of that we work our way up to certifying multithreaded, database-backed Web applications. Key verified components include a cooperative threading library and an implementation of a domain-specific language for XML processing. We have deployed our case-study system on mobile robots, where it interfaces with off-the-shelf components for sensing, actuation, and control., National Science Foundation (U.S.) (NSF grant CCF-1253229), United States. Defense Advanced Research Projects Agency (DARPA, agreement number FA8750-12-2-0293)

The World Wide Web has evolved gradually from a document delivery platform to an architecture for distributed programming. This largely unplanned evolution is apparent in the set of interconnected languages and protocols that any Web application must manage. This paper presents Ur/Web, a domain-specific, statically typed functional programming language with a much simpler model for programming modern Web applications. Ur/Web's model is unified, where programs in a single programming language are compiled to other "Web standards" languages as needed; modular, supporting novel kinds of encapsulation of Web-specific state; and exposes simple concurrency, where programmers can reason about distributed, multithreaded applications via a mix of transactions and cooperative preemption. We give a tutorial introduction to the main features of Ur/Web, formalize the basic programming model with operational semantics, and discuss the language implementation and the production Web applications that use it., National Science Foundation (U.S.) (Grant CCF-1217501)

There are many domain libraries, but despite the performance benefits of compilation, domain-specific languages are comparatively rare due to the high cost of implementing an optimizing compiler. We propose commensal compilation, a new strategy for compiling embedded domain-specific languages by reusing the massive investment in modern language virtual machine platforms. Commensal compilers use the host language's front-end, use host platform APIs that enable back-end optimizations by the host platform JIT, and use an autotuner for optimization selection. The cost of implementing a commensal compiler is only the cost of implementing the domain-specific optimizations. We demonstrate the concept by implementing a commensal compiler for the stream programming language StreamJIT atop the Java platform. Our compiler achieves performance 2.8 times better than the StreamIt native code (via GCC) compiler with considerably less implementation effort., United States. Dept. of Energy. Office of Science (X-Stack Award DE-SC0008923), Intel Corporation (Science and Technology Center for Big Data), SMART3 Graduate Fellowship

We report on the design and implementation of an extensible programming language and its intrinsic support for formal verification. Our language is targeted at low-level programming of infrastructure like operating systems and runtime systems. It is based on a cross-platform core combining characteristics of assembly languages and compiler intermediate languages. From this foundation, we take literally the saying that C is a "macro assembly language": we introduce an expressive notion of certified low-level macros, sufficient to build up the usual features of C and beyond as macros with no special support in the core. Furthermore, our macros have integrated support for strongest postcondition calculation and verification condition generation, so that we can provide a high-productivity formal verification environment within Coq for programs composed from any combination of macros. Our macro interface is expressive enough to support features that low-level programs usually only access through external tools with no formal guarantees, such as declarative parsing or SQL-inspired querying. The abstraction level of these macros only imposes a compile-time cost, via the execution of functional Coq programs that compute programs in our intermediate language; but the run-time cost is not substantially greater than for more conventional C code. We describe our experiences constructing a full C-like language stack using macros, with some experiments on the verifiability and performance of individual programs running on that stack., United States. Defense Advanced Research Projects Agency (Agreement FA8750-12-2-0293))

We present Bolt, a novel system for escaping from infinite and long-running loops. Directed by a user, Bolt can attach to a running process and determine if the program is executing an infinite loop. If so, Bolt can deploy multiple strategies to escape the loop, restore the responsiveness of the program, and enable the program to deliver useful output. Bolt operates on stripped x86 and x64 binaries, dynamically attaches and detaches to and from the program as needed, and dynamically detects loops and creates program state checkpoints to enable exploration of different escape strategies. Bolt can detect and escape from loops in off-the-shelf software, without available source code, and with no overhead in standard production use. Copyright © 2012 ACM., NSF (Grant CCF-0811397), NSF (Grant CCF-0905244), NSF (IIS-0835652), NSF (Grant CCF-1036241), DARPA (Grant FA8650-11-C-7192), DARPA (Grant FA8750-12-2-0110), United States Department of Energy (Grant DE-SC0005288)

A technique for synthesizing iterators from declarative abstraction functions written in a relational logic specification language is described. The logic includes a transitive closure operator that makes it convenient for expressing reachability queries on linked data structures. Some optimizations, including tuple elimination, iterator flattening, and traversal state reduction, are used to improve performance of the generated iterators. A case study demonstrates that most of the iterators in the widely used JDK Collections classes can be replaced with code synthesized from declarative abstraction functions. These synthesized iterators perform competitively with the hand-written originals. In a user study the synthesized iterators always passed more test cases than the hand-written ones, were almost always as efficient, usually took less programmer effort, and were the qualitative preference of all participants who provided free-form comments.

We describe an approach for synthesizing data representations for concurrent programs. Our compiler takes as input a program written using concurrent relations and synthesizes a representation of the relations as sets of cooperating data structures as well as the placement and acquisition of locks to synchronize concurrent access to those data structures. The resulting code is correct by construction: individual relational operations are implemented correctly and the aggregate set of operations is serializable and deadlock free. The relational specification also permits a high-level optimizer to choose the best performing of many possible legal data representations and locking strategies, which we demonstrate with an experiment autotuning a graph benchmark.

Software developers commonly exploit multicore processors by building multithreaded software in which all threads of an application share a single address space. This shared address space has a cost: kernel virtual memory operations such as handling soft page faults, growing the address space, mapping files, etc. can limit the scalability of these applications. In widely-used operating systems, all of these operations are synchronized by a single per-process lock. This paper contributes a new design for increasing the concurrency of kernel operations on a shared address space by exploiting read-copy-update (RCU) so that soft page faults can both run in parallel with operations that mutate the same address space and avoid contending with other page faults on shared cache lines. To enable such parallelism, this paper also introduces an RCU-based binary balanced tree for storing memory mappings. An experimental evaluation using three multithreaded applications shows performance improvements on 80 cores ranging from 1.7x to 3.4x for an implementation of this design in the Linux 2.6.37 kernel. The RCU-based binary tree enables soft page faults to run at a constant cost with an increasing number of cores,suggesting that the design will scale well beyond 80 cores., Quanta Computer (Firm), National Science Foundation (U.S.). (Award 0834415), National Science Foundation (U.S.). (Award 0915164)

Shadow memory is used by dynamic program analysis tools to store metadata for tracking properties of application memory. The efficiency of mapping between application memory and shadow memory has substantial impact on the overall performance of such analysis tools. However, traditional memory mapping schemes that work well on 32-bit architectures cannot easily port to 64-bit architectures due to the much larger 64-bit address space. This paper presents EMS64, an efficient memory shadowing scheme for 64-bit architectures. By taking advantage of application reference locality and unused regions in the 64-bit address space, EMS64 provides a fast and flexible memory mapping scheme without relying on any underlying platform features or requiring any specific shadow memory size. Our experiments show that EMS64 is able to reduce the runtime shadow memory translation overhead to 81% on average, which almost halves the overhead of the fastest 64-bit shadow memory system we are aware of.

Helper locks allow programs with large parallel critical sections, called parallel regions, to execute more efficiently by enlisting processors that might otherwise be waiting on the helper lock to aid in the execution of the parallel region. Suppose that a processor p is executing a parallel region A after having acquired the lock L protecting A. If another processor p′ tries to acquire L, then instead of blocking and waiting for p to complete A, processor p′ joins p to help it complete A. Additional processors not blocked on L may also help to execute A. The HELPER runtime system can execute fork-join computations augmented with helper locks and parallel regions. HELPER supports the unbounded nesting of parallel regions.We provide theoretical completion-time and space-usage bounds for a design of HELPER based on work stealing. Specifically, let V be the number of parallel regions in a computation, let T1 [T subscript 1] be its work, and let eT¥ [~T subscript infinity symbol] be its “aggregate span” — the sum of the spans (critical-path lengths) of all its parallel regions. We prove that HELPER completes the computation in expected time O(T1/P+ eT¥+PV)[O (T subscript 1 / P plus ~T subscript infinity symbol plus PV] on P processors. This bound indicates that programs with a small number of highly parallel critical sections can attain linear speedup. For the space bound, we prove that HELPER completes a program using only O(PeS1)[O (P ~S subscript 1) stack space, where eS1 [~S subscript 1] is the sum, over all regions, of the stack space used by each region in a serial execution. Finally, we describe a prototype of HELPER implemented by modifying the Cilk multithreaded runtime system. We used this prototype to implement a concurrent hash table with a resize operation protected by a helper lock., National Science Foundation (U.S.) (Grant CNS-0540248) (Grant CNS-0615215)

We present an integrated proof language for guiding the actions of multiple reasoning systems as they work together to prove complex correctness properties of imperative programs. The language operates in the context of a program verification system that uses multiple reasoning systems to discharge generated proof obligations. It is designed to 1) enable developers to resolve key choice points in complex program correctness proofs, thereby enabling automated reasoning systems to successfully prove the desired correctness properties; 2) allow developers to identify key lemmas for the reasoning systems to prove, thereby guiding the reasoning systems to find an effective proof decomposition; 3) enable multiple reasoning systems to work together productively to prove a single correctness property by providing a mechanism that developers can use to divide the property into lemmas, each of which is suitable for a different reasoning system; and 4) enable developers to identify specific lemmas that the reasoning systems should use when attempting to prove other lemmas or correctness properties, thereby appropriately confining the search space so that the reasoning systems can find a proof in an acceptable amount of time. The language includes a rich set of declarative proof constructs that enables developers to direct the reasoning systems as little or as much as they desire. Because the declarative proof statements are embedded into the program as specialized comments, they also serve as verified documentation and are a natural extension of the assertion mechanism found in most program verification systems. We have implemented our integrated proof language in the context of a program verification system for Java and used the resulting system to verify a collection of linked data structure implementations. Our experience indicates that our proof language makes it possible to successfully prove complex program correctness properties that are otherwise beyond the reach of automated reasoning systems., Swiss National Science Foundation, National Science Foundation, DARPA