Browse Prior Art Database

Logically Shared Memory

IP.com Disclosure Number: IPCOM000108662D
Original Publication Date: 1992-Jun-01
Included in the Prior Art Database: 2005-Mar-22
Document File: 6 page(s) / 210K

Publishing Venue

IBM

Related People

Ammann, EM: AUTHOR [+2]

Abstract

This article describes an efficient LSM implementation with a distributed real memory. Basic Concept of Logically Shared Memory (LSM)

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 42% of the total text.

Logically Shared Memory

       This article describes an efficient LSM implementation
with a distributed real memory.
Basic Concept of Logically Shared Memory (LSM)

      LSM is a parallel processing machine architecture which gives
programmers the impression of a shared memory, although in reality
the memory may be distributed to multiple processors.  There are
different methods of achieving the impression of LSM.  With the
proposed machine architecture, LSM is only explicitly accessible.
This means that before a data field in LSM can be accessed, an access
right must be requested by a LOCK instruction, and the type of access
(READ or RITE) must be specified.  In addition to controlling access
to the LSM, the LOCK instruction copies the data field from the LSM
to the processors' private memory.  After completion of access, the
programmer issues an UNLOCK instruction.  If the access was a WRITE,
then the UNLOCK instruction implies copying back the data field from
the processors' private memory to the LSM.  This way, the programmer
has the impression of a shared memory; the data movement from the
possibly distributed memory to the processors' private memory occurs
implicitly in response to the LOCK, UNLOCK instructions (Fig. 1).

      In addition to the LOCK and UNLOCK instructions, it may be
useful to support further memory access instructions, such as
COMPARE-AND-SWAP, READ-UNSAFE, and WRITE-UNSAFE.
Implementation Alternatives

      One of the major advantages of this LSM architecture is that it
allows a large variety of actual hardware configurations.  Of primary
interest are alternatives with respect to the residency of the LSM.
Some major alternatives are:
      1.   LSM residing on shared disk
           This alternative would result in a very simple, but
relatively slow, implementation of the LSM.
      2.   LSM residing in really shared memory
           With tightly coupled multiprocessing or multiple virtual
machine support, the LSM can be easily realized in the shared memory.
      3.   LSM residing in special processor within processor cluster
           Within a cluster of parallel processors, the LSM can be
realized by assigning it to a specific processor in the cluster.  All
access instructions of the LSM (LOCK, UNLOCK, COMPARE-AND-SWAP,
READ-UNSAFE) have to be sent to the appropriate processor.
      4.   LSM distributed over processor cluster
           Within a cluster of parallel processors, the LSM can be
realized by distributing it over all processors.  The LSM access
instructions (LOCK, UNLOCK, COMPARE-AND-SWAP, READ- UNSAFE) have to
be sent to the appropriate processor.
Proposed Implementation of LSM

      This section proposes an implementation of LSM in a
distributed-memory multiprocessor system with microcoded architecture
for interprocessor communication.

      Space for LSM and its handling logic is reserved i...