Browse Prior Art Database

Extensions to Round-CRMA: Reservation Oversampling, Service Round Monitoring and Adaptive Reservation Threshold

IP.com Disclosure Number: IPCOM000122287D
Original Publication Date: 1991-Nov-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 3 page(s) / 145K

Publishing Venue

IBM

Related People

Lemppenau, WW: AUTHOR [+2]

Abstract

The MAC protocol of Round-CRMA (R-CRMA) [*] specifies four topologies (single ring, dual ring, folded bus, dual bus) and three modes of multiplexing (insertion, cell, pure). Improvements in several respects are described: (1) an equal and position independent opportunity to make reservations for a node's newest workload, (2) a reduction of delays in general and particularly the worst-case access delay, (3) an increase in system throughput and bandwidth utilization, (4) improved protocol robustness, and (5) clear functional partitioning between nodes and scheduler with respect to commands.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 43% of the total text.

Extensions to Round-CRMA: Reservation Oversampling, Service Round
Monitoring and Adaptive Reservation Threshold

      The MAC protocol of Round-CRMA (R-CRMA) [*] specifies
four topologies (single ring, dual ring, folded bus, dual bus) and
three modes of multiplexing (insertion, cell, pure). Improvements in
several respects are described: (1) an equal and position independent
opportunity to make reservations for a node's newest workload, (2) a
reduction of delays in general and particularly the worst-case access
delay, (3) an increase in system throughput and bandwidth
utilization, (4) improved protocol robustness, and (5) clear
functional partitioning between nodes and scheduler with respect to
commands.

      The focus is on a single ring topology and cell multiplexing.
The principles apply however also to other topologies and
multiplexing modes.  In R-CRMA, a reserve command is issued by the
scheduler immediately after the confirm command.  Nodes receiving the
confirm command determine their confirmed number of slots and let the
modified confirm command pass to their downstream neighbor. In
contrast, a node always holds the reserve command until all its
confirmed slot transmissions have taken place. Thereafter, requests
for the newest workload are included into the reserve command which
is then released. In case a node has no confirmations at the time
instant the reserve command is received, pending reservation requests
are added to the command which is then relayed immediately.  As the
reserve command finally returns to the scheduler, it signals the end
of the service round.  Now, a new round starts by processing the
reserve command and issuing the confirm and reserve command in this
order.  The total number of slots the scheduler confirms to the nodes
is limited by a threshold H.

      Characteristic for this mode of operation is that (1) owing to
reserve command holding, the nodes at the beginning of the ring have
not the same opportunity to include reservation requests for their
newest workload as have nodes at the end of the ring, that (2) in
average the network is not utilized for one network latency L between
two service rounds, that (3) spatial reuse decreases the length of a
service round (equal to H for heavy load) so that average network
utilization given by H/(H+L) also decreases.

      The following three modifications improve R-CRMA:
Oversampling

      The scheduler lets the reserve command continuously circulate,
whereby the command is not held by the nodes. The end of the service
round is now indicated by a flag set in the header of a slot.  This
end-of-round flag is issued by the scheduler at the begin of a
service round and is delayed by nodes still having slot
confirmations.  Thus, nodes never hold the reserve command and
include requests for the newest workload each time the reserve
command passes.  Upon return, the scheduler copies the contents of
the received reserve command in its comm...