ABSTRACT
With cloud service becoming more popular, low-latency communication is required between servers in a data center. Low-latency node-to-node or application-to-application notification can be achieved in a NUMA [1] (Non-Uniform Memory Access) system, but requires dedicated, special-purpose network infrastructure. However, IP networks are commonly used in data centers. Using a custom FPGA-based NIC, we create NUMA-like behavior using hardware-generated IP packets which can be transmitted over commodity Ethernet switches and IP routers. We demonstrate ping-pong acknowledgments between two PCs equipped with our IP-NUMA boards. Our IP-NUMA implementation exhibits latency as much as ten times lower than software using Berkeley sockets over a consumer-grade Ethernet switch. iSCSI initiator-target communications and transaction-based distributed software systems will benefit from the reduced latency.
- William J. Bolosky, Michael L. Scott, Robert P. Fitzgerald, Robert J. Fowler, and Alan L. Cox. NUMA policies and their relation to memory architecture. SIGARCH Comput. Archit. News, 19(2):212--221, April 1991. Google ScholarDigital Library
- W. Richard Stevens. UNIX Network Programming. Prentice-Hall, Inc., Upper Saddle River, NJ, USA, 1990. Google ScholarDigital Library
- Lattice Semiconductor Corporation. LatticeECP3 Versa Development Kit, 2013.Google Scholar
- Intel Corporation. PAUSE • Spin Loop Hint. In Intel 64 and IA-32 Architectures Software Developer • s Manual, volume 2B, pages 4--57, 2014.Google Scholar
- M. Eisler. XDR: External Data Representation Standard. RFC 4506 (INTERNET STANDARD), May 2006.Google Scholar
- G. Delp, A. Sethi, and D. Farber. An analysis of MemNet---an experiment in high-speed shared-memory local networking. In SIGCOMM '88: Symposium proceedings on Communications architectures and protocols, pages 165--174, New York, NY, USA, 1988. ACM. Google ScholarDigital Library
- Greg Finn. An integration of network communication with workstation architecture. ACM Computer Communication Review, October 1991. Google ScholarDigital Library
- David L. Tennenhouse, Joel F. Adam, David Carver, Henry H. Houh, Michael Ismert, Christopher Lindblad, William F. Stasior, David Wetherall, David R. Bacher, and Theresa Chang. The viewstation: A software-intensive approach to media processing and distribution. Multimedia Syst., 3(3):104--115, 1995. Google ScholarDigital Library
- Mark Hayter and Derek McAuley. The desk area network. SIGOPS Oper. Syst. Rev., 25(4):14--21, October 1991. Google ScholarDigital Library
- Rodney Van Meter, Steven Hotz, and Gregory Finn. Derived virtual devices: A secure distributed file system mechanism. In Ben Kobler, editor, Proc. Fifth NASA Goddard Conference on Mass Storage Systems and Technologies, September 1996.Google Scholar
- Intel Corporation. In Intel Data Direct I/O Technology (Intel DDIO): A Primer, 2012.Google Scholar
Index Terms
- IP-NUMA for low-latency communication
Recommendations
Dynamically-Allocated Multi-Queue Buffers for VLSI Communication Switches
Small n*n switches are key components of interconnection networks used in multiprocessors and multicomputers. The architecture of these n*n switches, particularly their internal buffers, is critical for achieving high-throughput low-latency ...
Low‐loss TCP/IP header compression for wireless networks
AbstractWireless is becoming a popular way to connect mobile computers to the Internet and other networks. The bandwidth of wireless links will probably always be limited due to properties of the physical medium and regulatory limits on the use of ...
Small scale multiprocessor soft IP (SSM IP): single FPGA chip area and performance evaluation
FPGA '09: Proceedings of the ACM/SIGDA international symposium on Field programmable gate arraysFuture generation multiprocessor system on chip (MPSOC) will be based on hundreds of processors connected through network on chips. One of the challenges is to tackle the design productivity required to reach this goal. We propose a NOC based small ...
Comments