Next:
Introduction
Up:
The Synergy Between Non-blocking
Previous:
Contents
List of Figures
Pseudo-code definition of DCAS (
Double-Compare-and-Swap
)
Deletion from the middle of list, protected by DCAS and version number.
DCAS Implementation using
LL
/
SC
and
LLP
/
SCP
. Success or failure of
SC
(and thus of the DCAS operation) is returned in U1 or whatever general register holds the argument to
SC
. 1 denotes success, 0 failure. If the next instruction tries to read U1, the hardware interlocks (as it already does for
LL/SC
) if the result of
SC
is not already in U1.
Approximate instruction times of extra overhead to synchronize deletion from a priority queue. This overhead does not include the backoff computation.
Performance of several synchronization algorithms with local work = 20 and the number of processes per processor = 1
Performance of several synchronization algorithms with local work = 2000 and the number of processes per processor = 1
Performance of several synchronization algorithms with local work = 2000 and the number of processes per processor = 3
Michael Greenwald
Thu Sep 19 12:18:20 PDT 1996
