Citation Context

...to simple shared objects. Research studies have been conducted that support the belief that simple objects are the predominate form of data sharing in many applications. For example, Tsigas and Zhang =-=[16, 17]-=- analyzed shared-memory applications from the SPLASH-2 benchmark suite [19] and Spark98 kernels suite [13] and found that almost all synchronization in these applications is for simple data structures...

Citation Context

... the nesting depth and the distribution of lock requests,as it relies on measuring synthetic tasks. Other studies have assessed the impact of lock-free synchronizationon large scientific applications =-=[15, 16]-=-. Unfortunately, these benchmarks are mostly concerned with overall per-formance and do not reveal the nature of individual critical sections. In this paper, we seek to provide additionaldata points o...

Citation Context

...on is guaranteed to finish in a bounded number of its own steps, regardless of the actions of concurrent operations. Non-blocking algorithms have been shown to have significant impact in applications =-=[11,12]-=-, and there is also a library, NOBLE [13], containing many implementations of non-blocking data structures. One of the most common synchronization primitives used in lock-free synchronization is the C...

Citation Context

...cts [4, 10] for this model. The wait-free and t-resilient objects can be deployed as building blocks in parallel programming to help parallel applications tolerate crash failures and gain performance =-=[13, 15, 19, 20]-=-. We observe that due to SIMD architecture each SIMD core with M hardware threads can read/write M memory locations in one atomic step. Using the M-register read/write operations we construct a wait-f...

Citation Context

...nments in the domain of concurrent programming. In the literature a number of efforts have been made to evaluate such methods through micro benchmarks [1], [9], [10] as well as macro benchmarks [11], =-=[12]-=-, [13]. These benchmarks try to rank synchronization constructs by measuring their potential for high throughput and also examine a subspace of the parameters that we examine in this paper. Evaluating...

Citation Context

...l to many computer applications, including scheduling applications based on linear programming and applications that simulate physical systems. More information about Spark98 can be found in [10]. In =-=[2]-=-, we showed that the non-blocking version of Spark98 performs better than the lock-based version and also better than the reduction-based version of Spark98. In this section, we examine the reason tha...

Citation Context

...times cannot be too high assuming that there is an upper bound on the number of elements. It has also been observed that in practice many operations on shared data objects have a short execution time =-=[4, 5]-=-. For these reasons, we will now consider the case where ∀i, j : Copi = C op j . We will consider Earliest-DeadlineFirst (EDF) scheduling [2]. At time t, it assigns priorities to jobs that have arrive...

Citation Context

...ads is to use lock primitives. Mutexes, semaphores, and critical sections are used to ensure that certain sections of code are executed in exclusion[1] To address these problems, researchers have proposed nonblocking algorithms for shared data objects. Nonblocking methods do not rely on mutual exclusion, thereby avoiding some of these inherent problems. Most non-blocking implementations guarantee that in any infinite execution, some pending operation completes within a finite number of steps. Nonblocking algorithms have been shown to be of big practical importance in practical applications [2][3] In the previous work we presented the concurrent access of Modified skip list with locking techniques[12] , as we have discussed the limitation due to locking method ,we present the lock free access of modified skip list data structure. This one is the initial efforts in this direction. 2. SKIP LIST AND MODIFIED SKIP LIST Skip-lists [4] are an increasingly important data structure for storing and retrieving ordered in-memory data. SkipLists have received little attention in the parallel computing world, in spite of their highly decentralized nature. This structure uses randomization and has a...

Citation Context

...tion (a) Original Data Structures 3 4 5 P e r c e n t o f C o r e U t i l i z a t i o n Seconds of Execution (b) TBB Data Structures 3 4 5 P e r c e n t o f C o r e U t i l i z a t i o n Seconds of Execution (c) Non blocking Data Structures Figure 11: Core Utilization Dedup utilization, with the TBB configuration having the highest average utilization, followed by the non-blocking configuration, and then by original configuration. This pattern reveals that in addition to completing computation faster, the non-blocking configuration also uses less processing power to do so. 10. Related Work In [19], Tsigas et al. present a performance study of integrating non-blocking containers into several key applications from the benchmark suites SPLASH-2 [20] and Spark98 [21]. The authors show that the use of non-blocking containers in large-scale high performance computing applications can lead to a performance increase by a factor of 24 when compared to the standard implementations. Their findings indicate that the use of non-blocking containers does not lead to performance loss in any use scenario. Further analysis by Tsigas et al. [19] reveals that large-scale scientific applications using non-...