C. C. Kuo, J. B. Carter, R. Kuramkote, and M. Swanson. As-coma: An adaptive hybrid shared memory architecture. In Proc. of International Conference on Parallel Processing, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and memory directory accesses in the nodes involved. These delays occur 1 on the critical path of serving the miss and are the major performance hurdle for CC NUMA multiprocessors. Thus, previous CC NUMA multiprocessor designs have focused on avoiding these operations or reducing their latencies [14, 24, 35, 39, 41]. Excessive remote memory operations in CC NUMA multiprocessors is commonly attributed to the inadequacy of the local cache hierarchy for supporting the large working sets found in practical scientific computations. Thus, a conventional strategy of previous CC NUMA designs arranges the memory ....

....architecture [31] a victim cache is employed to track the re fetches. Although the design requires significant modifications to the processor cache controller and bus protocol, simulation results demonstrate that it outperforms R NUMA. Another interesting design is the AS COMA (Adaptive S COMA) [24]. This approach initializes a page as a S COMA page and only switches to a CC NUMA page when the memory pressure is high, to prevent threshing. As mentioned above, traditional designs were mainly based on the premise that network operations are expensive and should be avoided or reduced. However, ....

C. C. Kuo, J. B. Carter, R. Kuramkote, and M. Swanson. As-coma: An adaptive hybrid shared memory architecture. In Proc. of International Conference on Parallel Processing, 1998.

....have proposed providing support for both protocols. For example the S3.mp [16] project at Sun Microsystems supports both a S COMA and CC NUMA protocols. HYBRID SCHEMES Given these diverse application requirements, hybrid schemes such as Reactive NUMA (RNUMA) 3] and Adaptive SCOMA (ASCOMA) [4] have been proposed. These techniques combine CC NUMA and S COMA to get the best of both with incrementally more hardware. These schemes have not yet been implemented in commercial systems. 3.2.5 REACTIVE NON UNIFORM MEMORY ACCESS (R NUMA) R NUMA dynamically reacts to program and system ....

C. Kuo, J. Carter, R. Kumarkote and M. Swanson. ASCOMA: An Adaptive Hybrid Shared Memory Architecture. In International Conference on Parallel Processing (ICPP'98) August 1998.

....and movement e.g. SGI Origin2000 [10] significantly limiting the applicability of page migration replication and diminishing the opportunity to reduce traffic. Alternatively, other DSMs incorporate aggressive hardware intensive techniques to implement fine grain (remote) memory caching [8,4,5,16] e.g. Sun WildFire [6] These designs are based on integrating the base CC NUMA protocol with Simple COMA [7] enabling a processor to store coherent remote memory blocks in main memory pages. Because main memory provides a much larger repository for remote caching, these designs potentially ....

....sharing degree. Because it allows read write caching of data in any sharer s local memory, R NUMA can reduce the overall number of capacity conflict misses in the system. The opportunity for reducing capacity conflict misses in R NUMA, however, highly depends on memory pressure on the node [8,5]. R NUMA can suffer from the problem of page fragmentation and requires large amount of memory to meet the working sets of applications. Practical implementations of R NUMA [6] also limit the page cache size to reduce hardware complexity and cost. Therefore, high memory pressure in some ....

[Article contains additional citation context not shown here]

Chen-Chi Kuo, John Carter, Ravindra Kuramkote, and Mark Swanson. Ascoma: An adaptive hybrid shared memory architecture. In Proceedings of the 1998 International Conference on Parallel Processing, August 1998.

....per node for each page [LL97] Each node can maintain a set of per page counters [FW97] Or, a page relocation counter can be attached to each cache set [MD98] The threshold value (K) which maximizes its performance could be changed depending on programs and memory pressure. Thus, the study in [KCKS98] proposes a scheme which dynamically adjusts the threshold value. It is also possible to divide the local memory into two parts: space always allocated for memory lines as in CC NUMA and (unallocated) space utilized for data replication as in DVSM. All of the schemes discussed so far could still ....

Chen-Chi Kuo, John Carter, Ravindra Kuramkote, and Mark Swanson. ASCOMA: An adaptive hybrid shared memory architecture. In Proceedings of the International Conference on Parallel Processing, pages 207--216, August 1998.

....performing essential kernel operations (i.e. those required by all architectures) KOVERHD: performing architecture specific kernel operations, such as remapping pages and handling relocation interrupts. U INSTR and U LC were satisfied, can be found in the technical report version of this paper [4]. We simulated the applications across a range of memory pressures between 10 and 90 . Only one result is shown for CC NUMA, since it is not affected by memory pressure. As can be seen in the graphs, the relative performance of the different architectures can vary dramatically as memory pressures ....

.... can lead to large performance gains (barnes, em3d, and radix) and (ii) applications in which minimal efforts to avoid thrashing are sufficient for handling high memory pressure (fft, ocean, and lu) The behavior of em3d shows the danger of focusing solely on reducing remote conflict misses [4] when designing a memory architecture. As shown in Figure 1, the performance of em3d on the hybrid architectures is quite sensitive to memory pressure. R NUMA outperforms CC NUMA until memory pressure approaches 70 , after which time its performance drops quickly. CC NUMA outperforms R NUMA by 5 ....

[Article contains additional citation context not shown here]

C.-C. Kuo, J. Carter, R. Kuramkote, and M. Swanson. ASCOMA: An adaptive hybrid shared memory architecture. Technical report, University of Utah - Computer Science Department, March 1998.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC