Impact of Configurable Network Features in ATOLL

To read the full-text of this research, you can request a copy directly from the authors.


ATOLL (Atomic Low Latency) is a new high speed interconnect solution for cluster computing. It is equipped with a fully integrated network on a chip, which is currently a unique feature. Therefore hostinterface, network interface and switch come as one, integrated, single solution. Moreover, the network interface card allows for dynamic reconfiguration during runtime. This includes dynamic profiling as well as dynamic (wormhole) routing. Last but not least, one hostinterface can be set up to allow for zero copy data transfers. This paper describes configurable network features in ATOLL and how applications and software libraries can benefit from them. Keywords: System Area Networks, Cluster Computing, Zero Copy, Profiling 1 Introduction The better cost/performance ratio of beowulf like cluster systems has led to using commodity off the shelf products for parallel processing. Besides costs, monolith MPP systems have the disadvantage to be a proprietary system in which software update...

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

ResearchGate has not been able to resolve any citations for this publication.
Full-text available
This report describes the design and implementation of a high performance MPI library using a zero copy message transfer primitive supported by PM. MPICH-PM consists of the MPICH implementation of the Message Passing Interface (MPI) standard, ported to the high-performance, communications library PM. The zero copy message transfer primitive requires a memory area pinned down to physical memory, which is a restricted resource quantity under a paging memory system. Allocation of pinned down memory by multiple simultaneous requests for sending and receiving without any control can cause deadlock. To avoid this deadlock, we have introduced: i) separate of control of send/receive pin-down memory areas to ensure that at least one send and receive may be concurrently processed, and ii) delayed queues to handle the postponed message passing operations which could not be pinned-down. Contents 1 Introduction 2 2 Design of the Zero Copy MPI Implementation 4 2.1 PM : : : : : : : ...
Full-text available
The problem with Gbit/s networks is to get the hardware performance into the applications. The most promising technique is a zero-copy protocol combined with a user-space communication subsystem that (a) gives the application direct access to the network interface and (b) avoids all buffering /copying. In this paper we examine the design space of user-space communication subsystems, especially how send and receive operations work and which communication semantics they imply. Furthermore, we propose a technique called page-exchange which avoids copy operations, is well defined, and has communication semantics equivalent to those of standard programming interfaces such as UNIX sockets, PVM, or MPI. Keywords: User-Space Communication, ZeroCopy Protocols, High-Speed Interconnects, HighPerformance Cluster Computing. 1 Introduction Modern high-speed networks, such as Myrinet [5], allow cost effective high-performance clusters to be built from commodity PCs and workstations. Traditional com...
The ATOLL approach for a fast and reliable System Area Network Third Intl
  • Jörg Kluge
  • Ulrich Brüning
  • Markus Fischer
  • Lars Rzymianowicz
  • Patrick Schulz
  • Mathias Waack
Jörg Kluge, Ulrich Brüning, Markus Fischer, Lars Rzymianowicz, Patrick Schulz and Mathias Waack. The ATOLL approach for a fast and reliable System Area Network Third Intl. Workshop on Advanced Parallel Processing Technologies (APPT'99) conference, October 19-21 1999, in Changsha, P.R. China.
The Design and Implementation of Zero Copy MPI using
  • O' Caroll
  • Tezuka
  • Ishikawa Hori
O' Caroll, Tezuka, Hori and Ishikawa. The Design and Implementation of Zero Copy MPI using..., In International Conference on Supercomputing '98, pages 243-250, July 1998.
MPICH-PM: Design and Implementation of Zero Copy MPI for PM
  • O'carroll
  • Hori Tezukua
O'Carroll, Tezukua, Hori, and Ishikawa. MPICH-PM: Design and Implementation of Zero Copy MPI for PM, Technical Report TR-97011, RWC, March 1998.