[RTEMS Project] #2273: One Step Towards Fine Grained Locking

[RTEMS trac]

#2273: One Step Towards Fine Grained Locking
-----------------------------+------------------------------
 Reporter:  sebastian.huber  |       Owner:  sebastian.huber
     Type:  enhancement      |      Status:  accepted
 Priority:  normal           |   Milestone:  4.11.1
Component:  SMP              |     Version:  4.11
 Severity:  normal           |  Resolution:
 Keywords:                   |
-----------------------------+------------------------------

Comment (by sebastian.huber):

 The new test [dc5e5f44485b8f9709da767f79595d2fa8aff74d/rtems] shows that
 the prototype implementation for the RTEMS events yields a significant
 improvement for SMP systems in case no blocking operations are involved.
 The test data is gathered on a 24 processor Freescale T4240 platform with
 the RTEMS version [6db5e650fb80824c4e9339dfe73976662245d449/rtems].

 The first plot shows the count of event send/receive operations for task
 sets of size 1 to 24 with the patch
 [7d6e94b12ab94f13d3e769702d50f91be7d5884b/rtems] reverted. The Giant lock
 limits the total operation count and no benefit from more processors is
 visible.

 [[Image(ticket:2273:test_with_giant_lock.png)]]

 The second plot shows the count of event send/receive operations for task
 sets of size 1 to 24. The sum of the requests per second increases in this
 test scenario with fine grained locking linearly with the task set
 cardinality up to the count of processors.

 [[Image(ticket:2273:Test_with_fgl.png)]]

 In addition the new implementation (fine grained locking) is more
 efficient on a task set size of one with 2439242 vs. 1367189 operations
 per second.

--
Ticket URL: <http://devel.rtems.org/ticket/2273#comment:7>
RTEMS Project <http://www.rtems.org/>
RTEMS Project