M. G. Katevenis, E. P. Markatos, G. Kalokairinos, and A. Dollas, Telegraphos: A substrate for high performance computing on workstation clusters, Journal of Parallel and Distributed Computing, 43 (1997), pp. 94--108.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....in certain situations [34, 26] they support only explicit operations; they do not require code instrumentation or observing the memory bus, or the network to provide global ordering guarantees. Thus, they can more likely be supported in commodity NIs. In fact, many modern communication systems [12, 20, 31, 24] and the recent Virtual Interface Architecture (VIA) standard [16] support some of these or similar types of operations. We find that: i) The proposed protocol extensions improve performance substantially for our suite of ten applications. Performance improves by about 38 on average for ....

....NI support [12, 20, 34, 26] In our implementation, non contiguous pieces of data are sent directly to remote data structures with separate messages, and are not packed into bigger messages or combined by scatter gather support. Many communication systems support this or similar type of operations [12, 20, 31, 24, 16]. We use the remote deposit mechanism in all our subsequent protocols to exchange small pieces of control information during barrier synchronization and to directly update other remote protocol data structures (e.g. page timestamps, barrier control information) In addition, there are two major ....

M. G. H. Katevenis, E. P. Markatos, G. Kalokerinos, and A. Dollas. Telegraphos: A substrate for high-performance computing on workstation clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, 15 June 1997.

....model attractive for application users on clusters as well, thus making it competitive with message passing in performance portability across high end multiprocessors and clusters. The most popular form of hardware support used so far is the propagation of fine grained writes to remote memories [6, 18, 11]. This support inspired the design of a family of new, home based protocols for page based software shared virtual memory (SVM) which di#er from earlier all software protocols not only in hardware support but also in the manner in which they propagate changes and solve the multiple writer ....

M. G. H. Katevenis, E. P. Markatos, G. Kalokerinos, and A. Dollas. Telegraphos: A substrate for high-performance computing on workstation clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, 15 June 1997.

.... much and what kind of limited hardware support is most effective in accelerating their performance and bringing it closer to that of hardware coherence (or message passing) The most popular form of hardware support used so far is the hardware propagation of fine grained writes to remote memories [4, 16, 9]. This support inspired the design of a family of new, home based protocols for page based software shared virtual memory (SVM) which differ from earlier all software protocols not only in hardware support but also in the manner in which they propagate changes and solve the multiple writer ....

M. G. H. Katevenis, E. P. Markatos, G. Kalokerinos, and A. Dollas. Telegraphos: A substrate for high-performance computing on workstation clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, 15 June 1997.

....model attractive for application users on clusters as well, thus making it competitive with message passing in performance portability across high end multiprocessors and clusters. The most popular form of hardware support used so far is the propagation of fine grained writes to remote memories [6, 18, 11]. This support inspired the design of a family of new, home based protocols for page based software shared virtual memory (SVM) which differ from earlier all software protocols not only in hardware support but also in the manner in which they propagate changes and solve the multiple writer ....

M. G. H. Katevenis, E. P. Markatos, G. Kalokerinos, and A. Dollas. Telegraphos: A substrate for high-performance computing on workstation clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, 15 June 1997.

No context found.

M. G. Katevenis, E. P. Markatos, G. Kalokairinos, and A. Dollas, Telegraphos: A substrate for high performance computing on workstation clusters, Journal of Parallel and Distributed Computing, 43 (1997), pp. 94--108.

No context found.

M. G.H. Katevenis, E. P.Markatos, G. Kalokairinos, and A. Dollas. Telegraphos: A Substrate for High Performance Computing on Workstation Clusters. Journal of Parallel and Distributed Computing, 43(2)- June 1997.

.... operating system call (normalized to the cost of an empty operating system call of our baseline system) We plot the kernel 1 More details about the computers used can be found in [7] 2 For some noticeable exceptions in the area of high speed communication for workstation clusters see [9] and [6]. 10 20 30 40 50 10 20 30 40 50 Kernel Entry Exit Latency Improvement (relative to baseline) SPEC Int95 ideal PENTIUM Linux SPARC Solaris ALPHA DigUNIX MIPS Irix Figure 1: Kernel Entry Exit latency. entry exit latency as a function of the processor speed (measured in SPECint95 ....

M. G.H. Katevenis, E. P. Markatos, G. Kalokairinos, and A. Dollas. Telegraphos: A Substrate for High Performance Computing on Workstation Clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, June 1997.

....arrival. In this paper we present the Remote Enqueue atomic operation, which allows user level processes to enqueue (short) data in remote queues that reside in various workstations in a cluster, with no need for prior synchronization. This operation was developed within the Telegraphos project [18], in order to provide a fast message arrival notification mechanism. The Telegraphos network interface provides user applications with the ability to read write remote memory locations, using regular load store instructions to remote memory addresses. Sending (short) messages in Telegraphos can be ....

Manolis G. H. Katevenis, Evangelos P. Markatos, George Kalokerinos, and Apostolos Dollas. Telegraphos: A Substrate for High-Performance Computing on Workstation Clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, June 1997.

....be developed over a high throughput, low latency network interface, like Myrinet [4] U net [31] Memory Channel [15] SCI [18] and ATM. Some Network interfaces have transparent hardware support for mirroring, which makes PERSEAS easier to implement. Such systems include PRAM [29] Telegraphos [21], and SHRIMP [1] PERSEAS is based on three main functions: ffl remote malloc: The remote malloc operation can be used to map physical memory from a remote node to the calling process s virtual address space. ffl remote free: Remote free is used to free an occupied remote main memory segment. ....

Manolis G.H. Katevenis, Evangelos P. Markatos, George Kalokairinos, and Apostolos Dollas. Telegraphos: A Substrate for High Performance Computing on Workstation Clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, June 1997.

No context found.

M. G. H. Katevenis, E. P. Markatos, G. Kalokerinos, and A. Dollas. Telegraphos: A substrate for high-performance computing on workstation clusters. Journal of Parallel and Distributed Computing, 43(2):94--108, 15 June 1997.

No context found.

M. Katevenis, E. Markatos and A. Dollas, "Telegraphos: a substrate for high performance computing on workstation clusters", J. Parallel Distrib. Comput. (1997) to appear.

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