As ever more computation shifts onto multicore architectures, it is increasingly critical to find effective ways of dealing with multithreaded performance bugs like true and false sharing. Previous approaches to fixing false sharing in unmanaged languages have had to resort to highly-invasive runtime program modification. We observe that managed language runtimes, with garbage collection and JIT code compilation, present unique opportunities to repair such bugs directly, mirroring the techniques used in manual repairs.
We present Remix, a modified version of the Oracle HotSpot JVM which can detect cache contention bugs and repair false sharing at runtime. Remix’s detection mechanism leverages recent performance counter improvements on Intel platforms, which allow for precise, unobtrusive monitoring of cache contention at the hardware level. Remix can detect and repair known false sharing issues in the LMAX Disruptor high-performance inter-thread messaging library and the Spring Reactor event-processing framework, automatically providing 1.5-2x speedups over unoptimized code and matching the performance of hand-optimization. Remix also finds a new false sharing bug in SPECjvm2008, and uncovers a true sharing bug in the HotSpot JVM that, when fixed, improves the performance of three NAS Parallel Benchmarks by 7-25x. Remix incurs no statistically-significant performance overhead on other benchmarks that do not exhibit cache contention, making Remix practical for always-on use.
Data races are one of the main culprits behind the complexity of multithreaded programming. Existing data race detectors require large amounts of metadata for each program variable to perform their analyses. The SlimFast system exploits the insight that there is a large amount of redundancy in this metadata: many program variables often have identical metadata state. By sharing metadata across variables, a large reduction in space usage can be achieved compared to the state-of-the-art FastTrack algorithm. SlimFast uses immutable metadata to safely support metadata sharing across threads while also accelerating concurrency control. SlimFast’s lossless metadata compression achieves these benefits while preserving soundness and completeness.
Across a range of Java benchmarks from the Java Grande, DaCapo and NAS Parallel Benchmark suites, SlimFast is able to reduce memory consumption by 1.76x on average, and up to 4.47x for some benchmarks, compared to FastTrack. By improving cache locality and simplifying concurrency control, SlimFast also accelerates data race detection by 1.24x on average, and up to 8.33x for some benchmarks, compared to FastTrack.
Contention for shared memory, in the forms of true sharing and false sharing, is a challenging performance bug to discover and to repair. Understanding cache contention requires global knowledge of the program's actual sharing behavior, and can even arise invisibly in the program due to the opaque decisions of the memory allocator. Previous schemes have focused only on false sharing, and impose significant performance penalties or require non-trivial alterations to the operating system or runtime system environment.
This paper presents the Light, Accurate Sharing dEtection and Repair (LASER) system, which leverages new performance counter capabilities available on Intel's Haswell architecture that identify the source of expensive cache coherence events. Using records of these events generated by the hardware, we build a system for online contention detection and repair that operates with low performance overhead and does not require any invasive program, compiler or operating system changes. Our experiments show that LASER imposes just 2% average runtime overhead on the Phoenix, Parsec and Splash2x benchmarks. LASER can automatically improve the performance of programs by up to 19% on commodity hardware.
We present Consequence, a deterministic multi-threading library. Consequence achieves deterministic execution via store buffering and strict ordering of synchronization operations. To ensure high performance under a wide variety of conditions, the ordering of synch operations is based on a deterministic clock, and store buffering is implemented using version-controlled memory.
Recent work on deterministic concurrency has proposed relaxing the consistency model beyond total store order (TSO). Through novel optimizations, Consequence achieves the same or better performance on the Phoenix, PARSEC and SPLASH-2 benchmark suites, while retaining TSO memory consistency. Across 19 benchmark programs, Consequence incurs a worst-case slowdown of 3.9× vs. pthreads, with 14 out of 19 programs at or below 2.5×. We believe this performance improvement takes parallel programming one step closer to "determinism by default".
Nondeterminism is a key challenge in developing multithreaded applications. Even with the same input, each execution of a multithreaded program may produce a different output. This behavior complicates debugging and limits one's ability to test for correctness. This non-reproducibility situation is aggravated on massively parallel architectures like graphics processing units (GPUs) with thousands of concurrent threads. We believe providing a deterministic environment to ease debugging and testing of GPU applications is essential to enable a broader class of software to use GPUs.
Many hardware and software techniques have been proposed for providing determinism on general-purpose multi-core processors. However, these techniques are designed for small numbers of threads. Scaling them to thousands of threads on a GPU is a major challenge. This paper proposes a scalable hardware mechanism, GPUDet, to provide determinism in GPU architectures. In this paper we characterize the existing deterministic and nondeterministic aspects of current GPU execution models, and we use these observations to inform GPUDet's design. For example, GPUDet leverages the inherent determinism of the SIMD hardware in GPUs to provide determinism within a wavefront at no cost. GPUDet also exploits the Z-Buffer Unit, an existing GPU hardware unit for graphics rendering, to allow parallel out-of-order memory writes to produce a deterministic output. Other optimizations in GPUDet include deterministic parallel execution of atomic operations and a workgroup-aware algorithm that eliminates unnecessary global synchronizations.
Our simulation results indicate that GPUDet incurs only 2× slowdown on average over a baseline nondeterministic architecture, with runtime overheads as low as 4% for compute-bound applications, despite running GPU kernels with thousands of threads. We also characterize the sources of overhead for deterministic execution on GPUs to provide insights for further optimizations.
Data-race freedom is a valuable safety property for multithreaded programs that helps with catching bugs, simplifying memory consistency model semantics, and verifying and enforcing both atomicity and determinism. Unfortunately, existing software-only race detectors are precise but slow; proposals with hardware support offer higher performance but are imprecise. Both precision and performance are necessary to achieve the many advantages always-on race detection could provide.
To resolve this trade-off, we propose RADISH, a hybrid hardware-software race detector that is always-on and fully precise. In RADISH, hardware caches a principled subset of the metadata necessary for race detection; this subset allows the vast majority of race checks to occur completely in hardware. A flexible software layer handles persistence of race detection metadata on cache evictions and occasional queries to this expanded set of metadata. We show that RADISH is correct by proving equivalence to a conventional happens-before race detector.
Our design has modest hardware complexity: caches are completely unmodified and we piggy-back on existing coherence messages but do not otherwise modify the protocol. RADISH can furthermore leverage type-safe languages to reduce overheads substantially. Our evaluation of a simulated 8-core RADISH processor using PARSEC benchmarks shows runtime overheads from negligible to 2x. Furthermore, RADISH outperforms the leading software-only race detector by 2x-37x.
Providing deterministic execution significantly simplifies the debugging, testing, replication, and deployment of multithreaded programs. Recent work has developed deterministic multiprocessor architectures as well as compiler and runtime systems that enforce determinism in current hardware. Such work has incidentally imposed strong memory-ordering properties. Historically, memory ordering has been relaxed in favor of higher performance in shared memory multiprocessors and, interestingly, determinism exacerbates the cost of strong memory ordering. Consequently, we argue that relaxed memory ordering is vital to achieving faster deterministic execution.
This paper introduces RCDC, a deterministic multiprocessor architecture that takes advantage of relaxed memory orderings to provide high-performance deterministic execution with low hardware complexity. RCDC has two key innovations: a hybrid HW/SW approach to enforcing determinism; and a new deterministic execution strategy that leverages data-race-free-based memory models (e.g., the models for Java and C++) to improve performance and scalability without sacrificing determinism, even in the presence of races. In our hybrid HW/SW approach, the only hardware mechanisms required are software-controlled store buffering and support for precise instruction counting; we do not require speculation. A runtime system uses these mechanisms to enforce determinism for arbitrary programs.
We evaluate RCDC using PARSEC benchmarks and show that relaxing memory ordering leads to performance and scalability close to nondeterministic execution without requiring any form of speculation. We also compare our new execution strategy to one based on TSO (total-store-ordering) and show that some applications benefit significantly from the extra relaxation. We also evaluate a software-only implementation of our new deterministic execution strategy.
The behavior of a multithreaded program does not depend only on its inputs. Scheduling, memory reordering, timing, and low-level hardware effects all introduce nondeterminism in the execution of multithreaded programs. This severely complicates many tasks, including debugging, testing, and automatic replication. In this work, we avoid these complications by eliminating their root cause: we develop a compiler and runtime system that runs arbitrary multithreaded C/C++ POSIX Threads programs deterministically.
A trivial non-performant approach to providing determinism is simply deterministically serializing execution. Instead, we present a compiler and runtime infrastructure that ensures determinism but resorts to serialization rarely, for handling interthread communication and synchronization. We develop two basic approaches, both of which are largely dynamic with performance improved by some static compiler optimizations. First, an ownership-based approach detects interthread communication via an evolving table that tracks ownership of memory regions by threads. Second, a buffering approach uses versioned memory and employs a deterministic commit protocol to make changes visible to other threads. While buffering has larger single-threaded overhead than ownership, it tends to scale better (serializing less often). A hybrid system sometimes performs and scales better than either approach individually.
Our implementation is based on the LLVM compiler infrastructure. It needs neither programmer annotations nor special hardware. Our empirical evaluation uses the PARSEC and SPLASH2 benchmarks and shows that our approach scales comparably to nondeterministic execution.
Writing shared-memory parallel programs is error-prone. Among the concurrency errors that programmers often face are atomicity violations, which are especially challenging. They happen when programmers make incorrect assumptions about atomicity and fail to enclose memory accesses that should occur atomically inside the same critical section. If these accesses happen to be interleaved with conflicting accesses from different threads, the program might behave incorrectly.
Recent architectural proposals arbitrarily group consecutive dynamic memory operations into atomic blocks to enforce memory ordering at a coarse grain. This provides what we call implicit atomicity, as the atomic blocks are not derived from explicit program annotations. In this paper, we make the fundamental observation that implicit atomicity probabilistically hides atomicity violations by reducing the number of interleaving opportunities between memory operations. We then propose Atom-Aid, which creates implicit atomic blocks intelligently instead of arbitrarily, dramatically reducing the probability that atomicity violations will manifest themselves. Atom-Aid is also able to report where atomicity violations might exist in the code, providing resilience and debuggability. We evaluate Atom-Aid using buggy code from applications including Apache, MySQL, and XMMS, showing that Atom-Aid virtually eliminates the manifestation of atomicity violations.
The C programming language is at least as well known for its absence of spatial memory safety guarantees (i.e., lack of bounds checking) as it is for its high performance. C's unchecked pointer arithmetic and array indexing allow simple programming mistakes to lead to erroneous executions, silent data corruption, and security vulnerabilities. Many prior proposals have tackled enforcing spatial safety in C programs by checking pointer and array accesses. However, existing software-only proposals have significant drawbacks that may prevent wide adoption, including: unacceptably high runtime overheads, lack of completeness, incompatible pointer representations, or need for non-trivial changes to existing C source code and compiler infrastructure.
Inspired by the promise of these software-only approaches, this paper proposes a hardware bounded pointer architectural primitive that supports cooperative hardware/software enforcement of spatial memory safety for C programs. This bounded pointer is a new hardware primitive datatype for pointers that leaves the standard C pointer representation intact, but augments it with bounds information maintained separately and invisibly by the hardware. The bounds are initialized by the software, and they are then propagated and enforced transparently by the hardware, which automatically checks a pointer's bounds before it is dereferenced. One mode of use requires instrumenting only malloc, which enables enforcement of per-allocation spatial safety for heap-allocated objects for existing binaries. When combined with simple intra-procedural compiler instrumentation, hardware bounded pointers enable a low-overhead approach for enforcing complete spatial memory safety in unmodified C programs.
Hardware transactional memory has great potential to simplify the creation of correct and efficient multithreaded programs, allowing programmers to exploit more effectively the soon-to-be-ubiquitous multi-core designs. Several recent proposals have extended the original bounded transactional memory to unbounded transactional memory, a crucial step toward transactions becoming a generalpurpose primitive. Unfortunately, supporting the concurrent execution of an unbounded number of unbounded transactions is challenging, and as a result, many proposed implementations are complex.
This paper explores a different approach. First, we introduce the permissions-only cache to extend the bound at which transactions overflow to allow the fast, bounded case to be used as frequently as possible. Second, we propose ONETM to simplify the implementation of unbounded transactional memory by bounding the concurrency of transactions that overflow the cache. These mechanisms work synergistically to provide a simple and fast unbounded transactional memory system.
The permissions-only cache efficiently maintains the coherence permissions--but not data--for blocks read or written transactionally that have been evicted from the processor's caches. By holding coherence permissions for these blocks, the regular cache coherence protocol can be used to detect transactional conflicts using only a few bits of on-chip storage per overflowed cache block. ONETM allows only one overflowed transaction at a time, relying on the permissions-only cache to ensure that overflow is infrequent. We present two implementations. In ONETM-Serialized, an overflowed transaction simply stalls all other threads in the application. In ONETM-Concurrent, non-overflowed transactions and non-transactional code can execute concurrently with the overflowed transaction, providing more concurrency while retaining ONETM's core simplifying assumption.
Nondeterminism is a key contributor to the difficulty of parallel programming. Many research projects have shown how to provide deterministic parallelism, but with unfortunate trade-offs. Deterministic execution enforces determinism for arbitrary programs but with significant runtime cost, while deterministic languages enforce determinism statically (without runtime overhead) but only for fork-join programs expressible in their static type systems.
MELD unifies these approaches. We explain the requirements for soundly integrating a deterministic language into a deterministic execution system, and describe a simple qualifier-based type checker that ensures isolation for code written in a deterministic language. We also extend MELD to incorporate nondeterministic operations without compromising the determinism of the rest of the program. Our experiments with benchmarks from the SPLASH2 and PARSEC suites show that a small number of annotations can accelerate the performance of deterministic versions of these programs by 2-6x.
This paper takes a critical look at the benefits provided by state-of-the-art deterministic execution techniques. Specifically, we look at four applications of deterministic execution: debugging, fault-tolerant replication, testing, and security. For each application, we discuss what an ideal system would provide, and then look at how deterministic systems compare to the ideal. Further, we discuss alternative approaches, not involving determinism, and we judge whether or not these alternatives are more suitable. Along the way, we identify open questions and suggest future work.
Ultimately, we find that there are competitive alternatives to determinism for debugging and replicating multithreaded programs; that determinism has high, though unproven, potential to improve testing; and that determinism has distinct security benefits in eliminating some covert timing channels. Furthermore, determinism is a unified solution for all four applications: this confers a distinct advantage over point solutions that do not compose well with one another.
The root of all (or most) evil in programming shared memory multiprocessors is that execution is not deterministic. Bugs are hard to find and non-repeatable. This makes debugging a nightmare and gives little assurance in the testing - there is no way to know how the program will behave in a different environment. Testing and debugging is already difficult with a single thread on uniprocessors. Pervasive parallel programs and chip multiprocessors will make this difficulty worse, and more widespread throughout our industry.
In this paper we make the case for fully deterministic shared memory multiprocessing. The idea is to make memory interleaving fully deterministic, in contrast to past approaches of simply replaying an execution based on a memory interleaving log. This causes the execution of a parallel program to be only function of its inputs, making a parallel program effectively behave like a sequential program. We show that determinism can be provided with reasonable performance cost and we also discuss the benefits. Finally, we propose and evaluate a range of implementation approaches.
Correctly synchronizing threads’ executions is one of the main difficulties of multithreaded programming. Incorrect synchronization causes many subtle concurrency errors such as data races and atomicity violations. To overcome the difficulty of synchronizing code, we propose Mostly Automatic Management of Atomicity (MAMA): a new programming and execution model, in which the programmer is relieved of specifying the program’s atomicity constraints and need only specify parallelism and ordering constraints. At runtime, MAMA conservatively over-approximates atomicity, leveraging the insight that a large atomic region provides the atomicity of a smaller atomic region it contains. MAMA’s conservatism is sometimes excessive, and can cause deadlocks. However, our experiments with PARSEC workloads and the memcached server show that deadlocks are rare and can be repaired through programmer annotations. Our system precisely guides the programmer to annotation sites and suggests over 80% of annotations automatically. With modest hardware support, MAMA incurs just 40% average slowdown and nearly identical performance scaling compared to native pthreads.
Contention for shared memory, in the forms of true sharing and false sharing, is a challenging performance bug to discover and to repair. Cache contention is often invisible at the source code level as it is a product of the opaque decisions of the memory allocator, and contention can appear or disappear across different hardware platforms.We describe the Low-overhead, Effective, and Non-intrusive Sharing detection (LENS) system, which leverages new performance counter capabilities available on Intel’s Haswell architecture that identify the source of expensive cache coherence events. Using records of these events generated by the hardware, we build a cache contention detector that operates with low performance overhead and does not require any invasive program, compiler or operating system changes. Our experiments show that LENS imposes just 3.6% average runtime overhead on the Phoenix, Parsec and Splash2X benchmarks, and accurately detects instances of true and false sharing in these workloads. We discuss several potential extensions for LENS to automatically repair contention in both managed and unmanaged code.
Correctly synchronizing the parallel execution of tasks remains one of the most difficult aspects of parallel programming. Without proper synchronization, many kinds of subtle concurrency errors can arise and cause a program to produce intermittently wrong results.I'll describe our initial steps towards a system that can automatically synchronize a set of programmer-specified, partially-independent parallel threads. Our current prototype system can infer much but not all of this synchronization, while providing a safety guarantee that a program either executes in a correctly atomic fashion or it deadlocks. Initial experiments indicate that our prototype can semi-automatically provide atomicity for a set of Java benchmarks while still allowing these benchmarks to execute in parallel.
Nondeterminism is a key complication in programming multicore systems. Previous approaches to coping with it have focused on replay or required the adoption of new languages, but my research has shown how to provide deterministic execution for arbitrary parallel programs written in existing languages. My work has examined how to build deterministic platforms using novel hardware, compilers, runtimes, and type systems. I will also discuss the many uses of determinism, from debugging, testing, and replicating multithreaded programs to using determinism to accelerate dynamic safety and security checks.
Nondeterminism is a key complication in programming multicore systems. It makes testing more difficult and less useful, since another run of the program can potentially introduce new behaviors. Nondeterminism also frustrates debugging efforts by making bugs hard to reproduce. Previous approaches to coping with nondeterminism in parallel programs have focused on recording an execution for subsequent replay, or required that programs be written in restrictive languages, but have not addressed the underlying nondeterminism of multicore systems in a direct way.
In this talk, I will show how to use novel hardware and software techniques to provide deterministic execution for arbitrary parallel programs written in today's languages. I've built a series of deterministic platforms, from new hardware architectures to compilers and language extensions, that show how the challenge of nondeterminism can be addressed across the computing stack. Hardware speculation, memory consistency relaxations, and hardware-software co-design all play key roles in improving the performance and simplicity of determinism. I'll also share my plans for future work, from leveraging determinism to accelerate safety and security checks, to new parallel computer architectures that enable a unified task+data parallelism abstraction.
Nondeterminism is one of the main reasons that parallel programming is so difficult. Bugs can vanish when programs are rerun or run under a debugger, thwarting attempts at their removal. Stress-testing is a common practice to flush out rare defects though it consumes extensive processing power and offers no real guarantees of discovering bugs. Deployment can similarly expose new issues that are difficult to reproduce. Finally, nondeterminism frustrates replicating multithreaded programs for fault-tolerance or performance as the replicas can diverge silently. Determinism eliminates these problems, making debugging and replication possible and making testing more valuable and efficient.
Previous efforts at providing determinism required programs to be (re)written in restrictive languages. In contrast to these language-level determinism schemes, this dissertation shows how execution-level determinism can be provided for arbitrary parallel programs, even programs that contain concurrency errors. First, we employ a hardware-based approach to provide determinism for unmodified binaries running on a deterministic multiprocessor system. Second, we show that memory consistency relaxations both enable a pure software-based implementation of execution-level determinism for arbitrary programs and also admit a simpler deterministic multiprocessor design. Finally, we describe a hybrid mechanism that integrates execution-level and language-level determinism techniques to provide determinism for arbitrary programs with higher performance than an execution-level approach alone.