-
Bug
-
Resolution: Fixed
-
P3
-
6
-
b51
-
generic
-
generic
Doug Lea writes:
Background: Bill Scherer and Michael Scott and I have been
discussing ways to use their "dual data structures".
http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/duals.html
Bill and I have been working on and off doing this for Exchanger and
SynchronousQueue. Exchanger is now ready.
...
The new version is consistently over twice as fast as old one
in the typical case of a single pair of threads using an exchanger,
and is almost arbitrarily faster when used by lots of threads.
For example, eleven times faster with 16 threads on 8way sparc.
It's also better-tested than the old one, since Bill has been
putting together various performance tests to collect data for
tech paper we'll probably write about these and related classes.
###@###.### 2005-05-02 00:37:05 GMT
Background: Bill Scherer and Michael Scott and I have been
discussing ways to use their "dual data structures".
http://www.cs.rochester.edu/u/scott/synchronization/pseudocode/duals.html
Bill and I have been working on and off doing this for Exchanger and
SynchronousQueue. Exchanger is now ready.
...
The new version is consistently over twice as fast as old one
in the typical case of a single pair of threads using an exchanger,
and is almost arbitrarily faster when used by lots of threads.
For example, eleven times faster with 16 threads on 8way sparc.
It's also better-tested than the old one, since Bill has been
putting together various performance tests to collect data for
tech paper we'll probably write about these and related classes.
###@###.### 2005-05-02 00:37:05 GMT
- relates to
-
JDK-6470111 Deadlock in Exchanger
- Resolved
-
JDK-6384064 SynchronousQueue put/take interrupt handling differs from timed poll
- Closed