C. Dubnicki, K. Li, and M. Mesarina. Network Interface Support for User-Level Buffer Management. In Parallel Computer Routing and Comm. Workshop, Univ. of Washington, May 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... In response to the incentives to reduce communication overhead in parallel computers several machines have been envisioned in which more of the network communication mechanisms are handled in hardware, e.g. the J machine [39] and the M machine [34] or in specialized network interfaces [43] 109] [55]. Such decreases in the overhead of a communication increase the relative importance of network topology. Yet contention based latency is a complicated function of topology and the communication pattern, and while topological considerations may provide performance optimizations, adding topology to ....

Cezary Dubnicki, Kai Li, and Malena Mesarina. Network Interface Support for User-Level Buffer Management. In First International Workshop, PCRCW'94, volume 853 of LNCS, pages 256--265, Seattle, Washington, USA, May 1994.

....traps and context switches fast, and to take advantage of common case behavior. The emergence of new multiprocessor network interfaces opens new possibilities for constructing network software. It is not always clear, however, which interface is most appropriate for which task. The SHRIMP project [5, 6, 9] at Princeton University supports user level communication between processes by mapping memory pages between virtual address spaces. This Virtual Memory Mapped network interface seems to have many advantages including flexible user level communication, and very low overhead to initiate data ....

....the same one used in the Intel Paragon. The key hardware component is the network interface board, which supports the virtual memory mapped communication (VMMC) model, to provide low overhead, protected, user level communication. For more details on the SHRIMP architecture the reader can consult [5, 6, 9]. VMMC is discussed in the next section. 3 Virtual Memory Mapped Communication Virtual memory mapped communication (VMMC) 10] was developed in response to the need for a basic multicomputer communication mechanism with extremely low latency and high bandwidth. These performance goals are ....

C. Dubnicki, K. Li, and M. Mesarina. Network interface support for user-level buffer management. Workshop on Parallel Computer Routing and Communications, Springer-Verlag, 1994.

....for the SHRIMP multicomputer which is being constructed at Princeton using Pentium PCs and an Intel Paragon routing network. In our first design, we explored how to do minimal modifications to the traditional DMAbased network interface design, while implementing virtual memory mapping in software [5]. Our design requires a system call to initiate outgoing data transfer, but its virtual memory mapped communication can reduce the send latency overhead by up to 78 . Received messages are transferred directly to memory, reducing the receive software overhead to only a few instructions in the ....

Cezary Dubnicki, Kai Li, and Malena Mesarina. Network interface support for user-level buffer management. In Workshop on Parallel Computer Routing and Communication. Springer-Verlag, 1994.

....in a multiuser environment. The implementation of the VMMC model requires close cooperation between software and hardware. This paper describes the software support for VMMC and presents its hardware requirements. We have implemented VMMC for two network interface designs in the SHRIMP project [1, 2, 4]. We achieved userto user latency of 4.8 sec and sustained bandwidth of 23 MB s, which is close to the peak hardware bandwidth. The software communication overhead is only a few user level instructions. In addition, the VMMC API is simple and its implementation requires minimal hardware and ....

....memory mapping completely in hardware. This approach provides fully protected, userlevel message passing, and it allows user programs to initiate an outgoing block data transfer with two userlevel instructions. Based on these network interfaces we built two implementations of VMMC. SHRIMP I [4], the first network interface designed by the SHRIMP team, uses the traditional DMA approach with added support for including destination physical address in the message header. On export, receive buffer pages are pinned in physical memory, so virtual memory mapping across nodes is implemented ....

C. Dubnicki, K. Li, and M. Mesarina. Network interface support for user-level buffer management. In Workshop on Parallel Computer Routing and Communication Workshop. Springer-Verlag, April 1994.

No context found.

C. Dubnicki, K. Li, and M. Mesarina. Network Interface Support for User-Level Buffer Management. In Parallel Computer Routing and Comm. Workshop, Univ. of Washington, May 1994.

No context found.

C. Dubnicki, K. Li, and M. Mesarina. Network Interface Support for User-Level Buffer Management. In Parallel Computer Routing and Comm. Workshop, Univ. of Washington, May 1994.

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