Branch mispredictions are inevitable in modern processors, but their high penalty in power and performance can be alleviated by selectively flushing the pipeline. In this paper, we present a practical technique to opportunistically determine the presence of correct path instructions in the processor pipeline and preserve those instructions instead of throwing them away. In our experiments, selective pipeline flush reduced the fetched instruction count by 7% (up to 26%), while increasing the performance by 2% (up to 15%).

Higher integration lowers total cost of ownership (TCO) in the data center by reducing equipment cost and lowering energy consumption. However, higher integration also makes it difficult to achieve guaranteed quality of service (QoS) for shared resources. Unlike many other resources, memory bandwidth cannot be finely controlled by software in existing systems. As a result, many systems running critical, bandwidth-sensitive applications remain underutilized to protect against bandwidth interference. In this paper, we propose a novel hardware architecture allowing practical, software-controlled partitioning of memory bandwidth. Proportionally Allocated Bandwidth at the Source and Target (PABST) precisely controls the bandwidth of applications by throttling request rates at the source and prioritizes requests at the target. We show that PABST is work conserving, such that excess bandwidth beyond the requested allocation will not go unused. For applications sensitive to memory latency, we pair PABST with a simple priority scheme at the memory controller. We show that when combined, the system is able to lower TCO by providing performance isolation across a wide range of workloads, even when co-located with memory-intensive background jobs.

Current flagship processors excel at extracting instruction-level-parallelism (ILP) by forming large instruction windows. Even then, extracting ILP is inherently limited by true data dependencies. Value prediction was proposed to address this limitation. Many challenges face value prediction, in this work we focus on two of them. Challenge #1: store instructions change the values in memory, rendering the values in the value predictor stale, and resulting in value mispredictions and a retraining penalty. Challenge #2: value mispredictions trigger costly pipeline flushes. To minimize the number of pipeline flushes, value predictors employ stringent, yet necessary, high confidence requirements to guarantee high prediction accuracy. Such requirements can negatively impact training time and coverage.

Virtualization has become a magic bullet to increase utilization, improve security, lower costs, and reduce management overheads. In many scenarios, the number of virtual machines consolidated onto a single processor has grown even faster than the number of hardware threads. This results in multiprogrammed virtualization where many virtual machines time-share a single processor core. Such fine-grain sharing comes at a cost; each time a virtual machine gets scheduled by the hypervisor, it effectively begins with a “cold” cache, since any cache blocks it accessed in the past have likely been evicted by other virtual machines. Recently, cache restoration prefetchers have been shown to reduce the cold cache effects caused by multiprogrammed virtualization. However, these prefetchers waste large amounts of bandwidth prefetching many useless blocks and writing and reading metadata to and from memory. Using block reuse based filtering, we enhance two existing cache restoration prefetchers and reduce their bandwidth overheads by 16% and 32%, with slowdowns of only 0.5% and 0.6%. We also propose a Region-Based Cache Restoration Prefetcher (RECAP), which groups cache blocks into coarse-grain regions of memory, and predicts which regions contain useful blocks to prefetch the next time the current virtual machine executes. Based on these predictions, and using a simple compression technique that exploits spatial locality, RECAP provides a robust prefetcher that improves performance by up to 42% for some applications and only uses 14% more bandwidth and 3.5% more power than a stride prefetcher, on average. Compared to other prefetchers designed for multiprogrammed virtualization, RECAP provides comparable performance while using an average of 12% to 27% less bandwidth, and reducing the energy-delay product of the L2 cache and main memory by 12% to 13% on average.

On the twentieth anniversary of the original publication [10], following ten years of intense activity in the research literature, hardware support for transactional memory (TM) has finally become a commercial reality, with HTM-enabled chips currently or soon-to-be available from many hardware vendors. In this paper we describe architectural support for TM added to a future version of the Power ISA™. Two imperatives drove the development: the desire to complement our weakly-consistent memory model with a more friendly interface to simplify the development and porting of multithreaded applications, and the need for robustness beyond that of some early implementations. In the process of commercializing the feature, we had to resolve some previously unexplored interactions between TM and existing features of the ISA, for example translation shootdown, interrupt handling, atomic read-modify-write primitives, and our weakly consistent memory model. We describe these interactions, the overall architecture, and discuss the motivation and rationale for our choices of architectural semantics, beyond what is typically found in reference manuals.

No abstract available.

The economics of server consolidation have led to the support of virtualization features in almost all server-class systems, with the related feature set being a subject of significant competition. While most systems allow for partitioning at the relatively coarse grain of a single core, some systems also support multiprogrammed virtualization, whereby a system can be more finely partitioned through time-sharing, down to a percentage of a core being allotted to a virtual machine. When multiple virtual machines share a single core however, performance can suffer due to the displacement of microarchitectural state. We introduce cache restoration, a hardware-based prefetching mechanism initiated by the underlying virtualization software when a virtual machine is being scheduled on a core, prefetching its working set and warming its initial environment. Through cycle-accurate simulation of a POWER7 system, we show that when applied to its private per-core L3 last-level cache, the warm cache translates into 20% on average performance improvement for a mixture of workloads on a highly partitioned core, compared to a virtualized server without cache restoration.

After many years of prefetching research, most commercially available systems support only two types of prefetching: software-directed prefetching and hardware-based prefetchers using simple sequential or stride-based prefetching algorithms. More sophisticated prefetching proposals, despite promises of improved performance, have not been adopted by industry. In this paper, we explore the efficacy of both hardware and software prefetching in the context of an IBM POWER6 commercial server. Using a variety of applications that have been compiled with an aggressively optimizing compiler to use software prefetching when appropriate, we perform the first study of a new runahead prefetching feature adopted by the POWER6 design, evaluating it in isolation and in conjunction with a conventional hardware-based sequential stream prefetcher and compiler-inserted software prefetching. We find that the POWER6 implementation of runahead prefetching is quite effective on many of the memory intensive applications studied; in isolation it improves performance as much as 36% and on average 10%. However, it outperforms the hardware-based stream prefetcher on only two of the benchmarks studied, and in those by a small margin. When used in conjunction with the conventional prefetching mechanisms, the runahead feature adds an additional 6% on average, and 39% in the best case (GemsFDTD).

Abstract Software transactional memory (STM) systems are an attractive environment to evaluate optimistic concurrency. We describe our experience of supporting and optimizing an STM system at both the managed runtime and compiler levels. We describe the design policies of our STM system and the statistics collected by the runtime to identify performance bottlenecks and guide tuning decisions. We present an initial work on supporting automatic instrumentation of the STM primitives for C/C++ and Java programs in the IBM XL compiler and J9 Java virtual machine. We evaluate and discuss the performance of several transactional programs running on our system. Copyright © 2008 John Wiley & Sons, Ltd.

Conventional relaxed memory ordering techniques follow a proactive model: at a synchronization point, a processor makes its own updates to memory available to other processors by executing a memory barrier instruction, ensuring that recent writes have been ordered with respect to other processors in the system. We show that this model leads to superfluous memory barriers in programs with acquire-release style synchronization, and present a combined hardware/software synchronization mechanism called conditional memory ordering (CMO) that reduces memory ordering overhead. CMO is demonstrated on a lock algorithm that identifies those dynamic lock/unlock operations for which memory ordering is unnecessary, and speculatively omits the associated memory ordering instructions. When ordering is required, this algorithm relies on a hardware mechanism for initiating a memory ordering operation on another processor. Based on evaluation using a software-only CMO prototype, we show that CMO avoids memory ordering operations for the vast majority of dynamic acquire and release operations across a set of multithreaded Java workloads, leading to significant speedups for many. However, performance improvements in the software prototype are hindered by the high cost of remote memory ordering. Using empirical data, we construct an analytical model demonstrating the benefits of a combined hardware-software implementation

Calling context profiles are used in many inter-procedural code optimizations and in overall program understanding. Unfortunately, the collection of profile information is highly intrusive due to the high frequency of method calls in most applications. Previously proposed calling-context profiling mechanisms consequently suffer from either low accuracy, high overhead, or both. We have developed a new approach for building the calling context tree at runtime, called adaptive bursting. By selectively inhibiting redundant profiling, this approach dramatically reduces overhead while preserving profile accuracy. We first demonstrate the drawbacks of previously proposed calling context profiling mechanisms. We show that a low-overhead solution using sampled stack-walking alone is less than 50% accurate, based on degree of overlap with a complete calling-context tree. We also show that a static bursting approach collects a highly accurate profile, but causes an unacceptable application slowdown. Our adaptive solution achieves 85% degree of overlap and provides an 88% hot-edge coverage when using a 0.1 hot-edge threshold, while dramatically reducing overhead compared to the static bursting approach.

Conventional out-of-order processors employ a multi-ported, fully-associative load queue to guarantee correct memory reference order both within a single thread of execution and across threads in a multiprocessor system. As improvements in process technology and pipelining lead to higher clock frequencies, scaling this complex structure to accommodate a larger number of in-flight loads becomes difficult if not impossible. Furthermore, each access to this complex structure consumes excessive amounts of energy. In this paper, we solve the associative load queue scalability problem by completely eliminating the associative load queue. Instead, data dependences and memory consistency constraints are enforced by simply reexecuting load instructions in program order prior to retirement. Using heuristics to filter the set of loads that must be re-executed, we show that our replay-based mechanism enables a simple, scalable, and energy-efficient FIFO load queue design with no associative lookup functionality, while sacrificing only a negligible amount of performance and cache bandwidth.

We present an algorithm for dynamically verifying that the execution of a multithreaded program is sequentially consistent. The algorithm uses a vector-timestamp logical time mechanism to construct and verify the acyclic nature of an execution's constraint graph.

Abstract We introduce a new technique for automated performance diagnosis, using the program's callgraph. We discuss our implementation of this diagnosis technique in the Paradyn Performance Consultant. Our implementation includes the new search strategy and new dynamic instrumentation to resolve pointer‐based dynamic call sites at run‐time. We compare the effectiveness of our new technique to the previous version of the Performance Consultant for several sequential and parallel applications. Our results show that the new search method performs its search while inserting dramatically less instrumentation into the application, resulting in reduced application perturbation and consequently a higher degree of diagnosis accuracy. Copyright © 2002 John Wiley & Sons, Ltd.

The use of the Java programming language for implementing server-side application logic is increasingly in popularity yet there is very little known about the architectural requirements of this emerging commercial workload. We present a detailed characterization of the Transaction Processing Council's TPC-W web benchmark, implemented in Java. The TPC-W benchmark is designed to exercise the web server and transaction processing system of a typical e-commerce web site. We have implemented TPC-W as a collection of Java servlets, and present an architectural study detailing the memory system and branch predictor behavior of the workload. We also evaluate the effectiveness of a coarse-grained multithreaded processor at increasing system throughput using TPC-W and other commercial workloads. We measure system throughput improvements from 8% to 41% for a two context processor, and 12% to 60% for a four context uniprocessor over a single-threaded uniprocessor despite decreased branch prediction accuracy and cache hit rates.

This paper explores the interaction of value prediction with thread-level parallelism techniques, including multithreading and multiprocessing, where correctness is defined by a memory consistency model. Value prediction subtly interacts with the memory consistency model by allowing data dependent instructions to be reordered We find that predicting a value and later verifying that the value eventually calculated is the same as the value predicted is not always sufficient. We present an example of a multithreaded pointer manipulation that can generate a surprising and erroneous result when value prediction is implemented without considering memory consistency correctness. We show that this problem can occur with real software, and we discuss how to apply existing techniques to eliminate the problem in both sequentially consistent systems and systems that obey relaxed memory consistency models.