Citation Context

...ng benchmark suites fall short of the presented requirements and must thus be considered unsuitable for evaluating CMP performance. SPLASH-2 SPLASH-2 is a suite composed of multi-threaded applications=-=[44]-=- and hence seems to be an ideal candidate to measure performance of CMPs. However, its program collection is skewed towards HPC and graphics programs. It thus does not include parallelization models s...

Citation Context

...section describes the simulation of LogTM and a baseline system using spin locks (Section 3.1) and compares them using a microbenchmark (Section 3.2) and parallel applications from the SPLASH-2 suite =-=[32]-=- (Section 3.3). 3.1. Target System & Simulation Model LogTM and the baseline system share the same basic SPARC/Solaris multiprocessor architecture, summarized in Table 3. Each system has 32 processors...

Citation Context

...ate of a remote virtual processor, for example TLB shootdowns and posting of an interrupt to a given virtual CPU. Overall, Disco is structured more like a highly tuned and scalable SPLASH application =-=[27]-=- than like a general-purpose operating system. 4.2.1 Virtual CPUs Like previous virtual machine monitors, Disco emulates the execution of the virtual CPU by using direct execution on the real CPU. To ...

Citation Context

...ark suite. Similarly, design of multiprocessor architectures, along with academic research in this area, have been heavily influenced by popular scientific and engineering benchmarks such as SPLASH-2 =-=[26]-=- and STREAMS [13], with only a handful of published architectural studies that have in some way tried to address issues specific to commercial workloads [3, 7, 9, 12, 14, 16, 20, 21, 24]. The lack of ...

Citation Context

...lementation. We primarily focus on characterizing the overhead of miss checks by presenting the static overheads for individual miss checks, the dynamic overheads for all of the SPLASH-2 applications =-=[22]-=-, and the frequency of instrumented accesses in these applications. In addition, we present preliminary parallel performance results for some of the SPLASH-2 applications running on a cluster of Alpha...

Citation Context

...a critical section may in fact not conflict, and such accesses do not require serialization. Two such examples are shown in Figure 1. Figure 1a shows an example from a multithreaded applicationsocean =-=[40]-=-. Since a store instruction (line 3) to a shared object is present, the lock is required. However, most dynamic executions do not perform the store operation and thus do not require the lock. Addition...

Citation Context

...efore a successive local lock re-acquire, thus reducing unfairness. 5.2 Applications We use barnes, cholesky, and mp3d from SPLASH [34] and radiosity, water-nsq, ocean-cont, and raytracesfrom SPLASH2 =-=[39]-=-. A locking version of mp3d is used to study the impact of TLR on a lock-intensive benchmark [16]. This version of mp3d does frequent synchronization to largely uncontended locks and lock access laten...

Citation Context

...s are actually implemented using data races. Examples include barriers, flag synchronization, producer-consumer queues, etc. Figure 2 gives two such examples from the SPLASH2 parallel benchmark suite =-=[40]-=-, in which user-defined barrier and flag synchronization are achieved via races on the shared variables b→synch and r, respectively. Furthermore, sometimes programmers intentionally choose to allow a ...

Citation Context

... Andrew Benchmark [Ousterhout90], while the MAB pmake is a slightly modified form of the MAB (described in [Rosenblum95a]). The other multiprocessor workloads are taken from the Splash benchmark suite=-=[Woo95]-=-. Those applications run with settings different from the default have their arguments shown. of these workloads, reflected by the large percentage of time spent in the translation cache, accounts for...