Citation Context

...d, using various techniques to minimize the software overhead for supporting the shared address space. For example, some systems require programmer annotations or explicit calls to access shared data =-=[2, 11]-=-. Another approach, called Shared Virtual Memory (SVM), uses the virtual memory hardware to detect access to data that is not available locally [4, 13, 12]. In most such systems, the granularity at wh...

Citation Context

...[17]. Both models take advantage of the association of data with locks, either explicitly (Entry Consistency) or implicitly (Scope Consistency), to reduce the protocol overhead. Both Orca [2] and CRL =-=[19]-=- are designed to implement distributed shared memory by maintaining coherence at object level instead of page level. These methods require specialized APIs, unlike the prototype systems presented in t...

Citation Context

...access checks in software. While the authors of Shasta used impressive compiler techniques to reduce the cost of these software access checks, MMP would reduce this cost further. The C region library =-=[JKW95]-=- also allows programmers to maintain coherence for arbitrary-sized data regions, and so would benefit from MMP’s ability to mark such data regions as inaccessible if they were owned exclusively by ano...

Citation Context

... if a process A reads a value written by another process B, then all subsequent operations by A must appear to occur after the write by B. Most DSM’s implement one of these relaxed consistency models =-=[11, 18, 21, 27]-=-, though some implement a fixed collection of consistency models [4], while others merely implement a collection of mechanisms on top of which users write their own DSM consistency policies [23, 26]. ...

Citation Context

...ural fit for in-memory computation: applications have no online access to intermediate state and often have to emulate shared memory access by joining multiple data streams. Distributed shared memory =-=[29, 32, 7, 17]-=- and tuple spaces [13] allow sharing of distributed in-memory state. However, their simple memory (or tuple) model makes it difficult for programmers to optimize for good application performance in a ...

Citation Context

... replaced by a queue of waiters at the home. Transparency of Shared Address Space Much work has been done on software shared memory using explicit program annotations to access the shared named space =-=[2,5,13,19,23]-=-. The difficulty of programming shared-memory machines using specific annotations, however, has hindered the widespread adoption of those systems. Hardware shared memory, while easier to program, is o...

Citation Context

...mory segments. The idea here is to allow groups of endpoints to keep specially-tagged endpoint virtual-memory segments cache-coherent. Similarly, it appears at first glance, that the C Region Library =-=[28]-=- could be easily incorporated into the exiting structures of the endpoints and associated memory segments. 4. Provide unreliable versions of the request and reply transport functions. In certain appli...

Citation Context

...s: a cache-line sized grain when using hardware shared memory, and a larger page-sized grain when using software shared memory. Other solutions to the grain size problem have explored variable grains =-=[4, 5]-=-. These solutions leverage application-specific knowledge to select the optimal grain. While this can be effective, it is difficult for a compiler and painful for a programmer. Multigrain shared memor...