Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A mechanism for address translation on network interfaces. In Proceedings of the Eighth International Conference Architectural Support for Programming Languages and Operating Systems ASPLOS, pages 193--203, San Jose, CA, Oct. 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....page faults at interrupt time is difficult at best because no context is available that can block waiting for the disk request. To avoid extensive modifications of the operating system kernel, the UTLB mechanism performs address translations and pins pages in advance under application control [21]. Applications manage a table of translations located at the DMA device. To implement protected address translation, applications have no access to the physical address of a mapping but specify DMA addresses using an index into the translation table. System calls are used to install mappings in ....

Y. Chen et al., "UTLB: A Mechanism for Address Translation on Network Interfaces," Proc. 8th Int'l Conf. Architectural Support for Programming Languages and Operating Systems (ASPLOS-VIII), ACM Press, New York, N.Y., 1998, pp. 193-204. 178

....localRequest Handles requests from the application to send receive data. remoteRequest Handles requests arriving on the network to send receive data. remoteReply Handles replies arriving on the network in response to a request sent. UTLB Translates virtual addresses to physical addresses [29]. ETLB Translates virtual addresses to physical addresses for exported regions. reliableSender A component of the retransmission protocol that accepts packets to be sent on the network. reliableReceiver A component of the retransmission protocol that accepts packets arriving on the network. ....

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A Mechanism for Address Translation on Network Interfaces. In Proceedings of the International Conference on Architectural Support for Programming Languages and Operating Systems, San Jose, California, October 1998.

....and compiler hinting [6, 36, 5] and even speculatively executing programs to get hints [10] There are also application interfaces, such as dynamic sets [51] and disk directed I O [27] that allow groups of requests to be specified collectively. Similarly, schemes like soft updates [17] and RIO [11] allow aggressive write back caching of data. However, in deciding what background accesses to do when, no information about what the disk could do efficiently is considered. 2.4 Previous Cross boundary Enhancements There do exist previous examples of cooperative enhancements of the form ....

Yuqun Chen, Angelos Bilas, Stefanos N. Damianakis, Cezary Dubnicki, and Kai Li. UTLB: a mechanism for address translation on network interfaces. Architectural Support for Programming Languages and Operating Systems (San Jose, CA, 3-7 October 1998.

....Channel features and the home node migration optimization. Section 4 covers related work, and Section 5 outlines our conclusions. 1 Most current commodity remote access networks have a limited remotely accessible memory space. Methods to eliminate this restriction are a focus of ongoing research [6, 26]. 3 2 Protocol Variants and Implementation Cashmere was designed for SMP clusters connected by a high performance network, specifically, Compaq s Memory Channel network [15] Earlier work [12, 14, 20, 21, 22, 25] on Cashmere and other systems has quantified the benefits of SMP nodes to SDSM ....

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A Mechanism for Address Translation on Network Interfaces. In Proceedings of the Eighth International Conference on Architectural Support for Programming Languages and Operating Systems, San Jose, CA, October 1998.

....and the home node migration and broadcast optimizations. Section 4 covers related work, and Section 5 outlines our conclusions. 1 Most current commodity remote access networks have a limited remotely accessible memory space. Methods to eliminate this restriction are a focus of ongoing research [7, 30]. 2. Protocol Variants and Implementation Cashmere was designed for SMP clusters connected by a high performance system area network, such as Compaq s Memory Channel [15] Earlier work on Cashmere [12, 28] and other systems [12, 14, 22, 23, 24] has quantified the benefits of SMP nodes to SDSM ....

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A Mechanism for Address Translation on Network Interfaces. In Proc. of the 8th Intl. Conf. on Architectural Support for Programming Languages and Operating Systems, pages 193--203, San Jose, CA, Oct. 1998.

....grow beyond the addressing limits of the network interface. 1 It also enables us to implement variants of Cashmere that employ 1 Most current commodity remote access networks have a limited remotely accessible memory space. Methods to eliminate this restriction are a focus of ongoing research [7, 30]. home node migration. These variants improve performance by as much as 67 , more than offsetting the advantage of using the network interface support in the base protocol. These results suggest that for systems of modest size (up to 8 nodes) low latency is much more important for SDSM ....

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A Mechanism for Address Translation on Network Interfaces. In Proceedings of the Eighth International Conference on Architectural Support for Programming Languages and Operating Systems, pages 193--203, San Jose, CA, October 1998.

No context found.

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A mechanism for address translation on network interfaces. In Proceedings of the Eighth International Conference Architectural Support for Programming Languages and Operating Systems ASPLOS, pages 193--203, San Jose, CA, Oct. 1998.

....the NIC. The simple solution of pre registering all I O bu#ers at application startup cannot be applied in database systems, since they use large caches and require large numbers of I O bu#ers. Given that we need to dynamically manage registered memory, previous work for user level communication [8, 4] has shown how the NIC can collaborate with host level software (either kernel or user level) to manage large amounts of host memory. These solutions have been evaluated in cases where the working set of registered memory is small. However, database systems use practically all available I O ....

....at low cost. Besides VI, specific examples of user level communication systems include Active Messages [9] BDM [19] Fast Messages [25] PM [27] U Net [3] and VMMC [12] Previous work in user level communication has also addressed the issue of dynamic memory registration and deregistration [8, 4]. These solutions target applications with small working sets for registered memory and require modifications either in the NIC firmware or the OS kernel. Using VI for databases: VI based interconnects have been used previously by database systems for purposes other than improving storage I O ....

Y. Chen, A. Bilas, S. N. Damianakis, C. Dubnicki, and K. Li. UTLB: A mechanism for address translation on network interfaces. In Proceedings of the Eighth International Conference Architectural Support for Programming Languages and Operating Systems ASPLOS, pages 193--203, San Jose, CA, Oct. 1998.

....support and the other requiring no special network interface support. To evaluate these protocols and investigate the design tradeo s of the network interface hardware, we implemented two fault tolerant PC clusters, one using SHRIMP network interfaces [14, 13] and the other using Myrinet [3, 9, 7] network interfaces. Both systems support virtual memory mapped communication. Our results with three transactionbased applications show that VMMC is a convenient and ef cient communication mechanism to support fast failover on clusters. With no special network interface support, our system ....

....mechanism is needed to transfer data eciently between the two application virtual address spaces (primary and backup) in order to replicate application data while minimizing interference with the computation. Among existing communication models, virtual memory mapped communication (VMMC) [14, 9, 7] is a good candidate because it provides direct data transfer between virtual address spaces. With the VMMC model, the receiver exports variable sized regions of contiguous virtual memory, called receive bu ers, with a set of permissions. Any other process, with proper permission, can import the ....

[Article contains additional citation context not shown here]

Y. Chen, S. Damianakis, C. Dubnicki, and K. Li. UTLB: A Mechanism for Address Translation on Network Interfaces. In ASPLOS, 1998.

....to implement. One technique is to rely on write combining to reduce the cost of PIO to acceptable levels. Use of DMA incurs certain setup costs, and also requires one to deal with issues of virtual physical address mapping and protection, since the DMA engine works with physical addresses only [21, 6]. Finally, data is transferred between the network interfaces using DMA engines on each network interface. The LANai processor has direct access to the network and it could perform the sends by copying data to control registers. This however, would tie the processor to the transfer, and it would ....

....issues. The Send Data and Recv Data rows refer to the mechanism used to transfer long messages from and to arbitrary locations in user memory. All libraries, except PM, use PIO to transfer short messages. WC stands for write combining and v2p trans for virtual to physical translation [21, 6]. Protection refers to how each system achieves transfers to and from arbitrary locations in virtual memory, without compromising system protection. Although BIP uses virtual to physical translation, it gives direct access to control information on the LANai, and thus does not provide protection. ....

Y. Chen, C. Dubnicki, S. Damianakis, A. Bilas, and K. Li. Utlb: A mechanism for address translation on network interfaces. In The 8th International Conference on Architectural Support for Programming Languages and Operating Systems, Oct. 1998.

No context found.

Y. Chen, S. Damianakis, C. Dubnicki, and K. Li. UTLB: A mechanism for address translation on network interfaces. In ASPLOS, 1998.

No context found.

Y. Chen, C. Dubnicki, S. Damianakis, A. Bilas and K. Li. "UTLB: A Mechanism for Address Translation on Network Interfaces." In Proceedings of the Eighth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS). 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