Intel Corporation. Pentium Processor Data Book, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of operating system pages, namely shared virtual memory (SVM) 54] This allows access control to be managed by the operating system via the virtual memory system. Communication and the coherence protocols are embedded in the page fault handlers. The specific cluster we use has Pentium Pro [36, 64] processors connected by a Myrinet [10] interconnect. The cost of communication and coherence in SVM systems are usually high because they are managed by software. The use of pages rather than cache blocks for communication and coherence in SVM systems can bring prefetching to hide latency and ....

Intel Corporation. Pentium Processor Data Book, 1993.

....implementation. Such experiments allow us to determine the best message size at which to switch protocols for optimal message passing performance. We conducted our experiments on a four node Shrimp multicomputer prototype. Each node of the Shrimp system is a 60 MHz Pentium based Xpress PC system [18, 14, 13], with 32 MB DRAM and a 256 kB second level cache. The nodes use an EISA [3] I O bus. The operating system on each node is Linux with a thin layer of software to provide virtual memorymapped communication [10] Results To measure latency and bandwidth for each message size, we performed 10; 000 ....

Intel Corporation. Pentium Processor Data Book, 1993.

....directly from the translation table. The third idea is to construct a fast user level lookup data structure. The user process has to keep track of the mapping between the translation table indices and the pinned virtual pages. The lookup table uses a standard two level page table architecture [21, 26]. It contains one entry for each virtual page. An entry can either be invalid or contain the index in the translation table where the physical address for this virtual page is stored. Only two memory references are required to obtain the UTLB index for a given virtual page address. 3.1 ....

Intel Corporation. Pentium Processor Data Book, 1993.

....Intel Pentium Xpress PC system [15] and the interconnect is an Intel Paragon routing backplane. These state of the art components allow SHRIMP to take advantage of the latest available technology at a fraction of the cost of current multicomputer systems. The Xpress PC consists of a Pentium CPU [28, 16] with a second level cache connected to DRAM memory modules and I O bus adapters (EISA [3] or PCI [25] via the Xpress memory bus. Memory can be cached as write through or write back on a per virtual page basis, as specified in process page tables. The caches snoop DMA transactions and ....

....context switch could happen at any time. This requires processes to use a single, atomic instruction to request a transfer and determine whether it was started. The current network interface supports deliberateupdate transfer initiation using the Pentium compareand exchange (CMPXCHG) instruction [16], which generates a read cycle followed by a write cycle if the value returned by the read matches the accumulator. To transfer n words of data starting from a mapped out base address, a process loads a source register with n, clears the accumulator, and issues a CMPXCHG instruction whose ....

Intel Corporation. Pentium Processor Data Book, 1993.

....an Intel Pentium Xpress PC system [13] and the interconnect is an Intel Paragon routing backplane. These state of the art components allow SHRIMP to take advantage of the latest available technology at a fraction of the cost of current multicomputer systems. The Xpress PC consists of a Pentium CPU [26, 14] with a second level cache connected to DRAM memory modules and I O bus adapters (EISA [3] or PCI [23] via the Xpress memory bus. Memory can be cached as write through or write back on a per virtual page basis, as specified in process page tables. The caches snoop DMA transactions and ....

....so a context switch could happen at any time. This requires processes to use a single, atomic instruction to start a transfer and determine whether it succeeded. The current network interface supports deliberateupdate transfer initiation using the compare andexchange (CMPXCHG) instruction [14], which generates a read cycle followed by a write cycle if the value returned by the read matches the accumulator. To transfer n words of data starting from a mapped out base address, the application loads a source register with n, and issues a CMPXCHG instruction whose destination address is the ....

Intel Corporation. Pentium Processor Data Book, 1993.

....work. We present the internal representation of the source assembly program, the format for the output graphs, the Shishya subsystem and the reorganization support provided. The Guru framework is currently available on the Linux platform with gcc support. The framework supports the Pentium i486[6] architecture. 4.1 Source program representation Figure 3 presents a schematic for the source program representation. The assembly source is stored in a hierarchical link list that begins with a list of files comprising the program. As depicted in figure 3 each file has a list of procedures. ....

Intel Corporation. Pentium processor data book. 1993.

....implementation. Such experiments allow us to determine the best message size at which to switch protocols for optimal message passing performance. We conducted our experiments on a four node Shrimp multicomputer prototype. Each node of the Shrimp system is a 60 MHz Pentium based Xpress PC system [18, 14, 13], with 32 MB DRAM and a 256 kB second level cache. The nodes use an EISA [3] I O bus. The operating system on each node is Linux with a thin layer of software to provide virtual memorymapped communication [10] Results To measure latency and bandwidth for each message size, we performed 10; 000 ....

Intel Corporation. Pentium Processor Data Book, 1993.

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