上午测试了下并发情况下的表现，测试场景：一个有20个节点，包括选择、顺序、并行路由的流程，所有节点都设置为自动执行，1000个线程并发启动案例，也就是这个流程同时有1000个案例在跑，全部跑完结果差强人意，比单线程慢了接近30倍。仔细调整了算法和加锁粒度，尽管整体性能有所提高，但是多线程和单线程间执行的差距仍然没有多大变化，性能瓶颈还是在核心的调度算法上，还需要分析下。测试程序如下：
package
net.rubyeye.insect.workflow.test;
import
java.util.ArrayList;
import
java.util.List;
import
java.util.concurrent.CyclicBarrier;
import
net.rubyeye.insect.workflow.Place;
import
net.rubyeye.insect.workflow.Token;
import
net.rubyeye.insect.workflow.Transition;
import
net.rubyeye.insect.workflow.WorkFlow;
import
net.rubyeye.insect.workflow.WorkFlowManager;
import
net.rubyeye.insect.workflow.basic.BasicWorkflowManager;
import
net.rubyeye.insect.workflow.comm.TransitionType;
import
net.rubyeye.insect.workflow.config.DefaultConfiguration;
import
junit.framework.TestCase;
public
class
CompositeProcessTest
extends
TestCase {
private
WorkFlowManager wm; WorkFlow composite;
private
CyclicBarrier barrier;
private
static
final
int
total
=
1000
; @Override
protected
void
setUp()
throws
Exception {
this
.barrier
=
new
CyclicBarrier(total
+
1
); wm
=
new
BasicWorkflowManager(); wm.setConfiguration(
new
DefaultConfiguration()); WorkFlow sequence
=
wm.getWorkFlow(
"
sequence
"
); WorkFlow concurrency
=
wm.getWorkFlow(
"
concurrency
"
); WorkFlow choose
=
wm.getWorkFlow(
"
choose
"
);
//
组合流程
composite
=
new
WorkFlow(); composite.setName(
"
composite
"
); composite.setId(
100
); wm.saveWorkFlow(composite);
//
修改开始结束节点的输入输出库所
sequence.getEnd().setType(TransitionType.NORMAL); sequence.getEnd().setOutputs(concurrency.getStart().getInputs()); concurrency.getEnd().setType(TransitionType.NORMAL); concurrency.getEnd().setOutputs(choose.getStart().getInputs()); composite.setStart(sequence.getStart()); composite.setEnd(choose.getEnd()); List
<
Transition
>
transitions
=
new
ArrayList
<
Transition
>
(); transitions.addAll(sequence.getTransitions()); transitions.addAll(concurrency.getTransitions()); transitions.addAll(choose.getTransitions()); composite.setTransitions(transitions); }
public
void
testConcurrencyCompositeProcesss()
throws
Exception {
for
(
int
i
=
0
; i
<
total; i
++
) {
new
FlowThread().start(); } barrier.await();
long
start
=
System.currentTimeMillis(); barrier.await();
long
end
=
System.currentTimeMillis(); System.out.println(
"
创建
"
+
total
+
"
个流程并发运行完毕\n花费时间:
"
+
(end
-
start)
/
1000.0
+
"
秒
"
);
for
(Transition transition : composite.getTransitions()) { System.out.println(transition.getName()
+
"
"
+
transition.isEnable());
for
(Place place : transition.getOutputs()) { System.out.println(
"
place
"
+
place.getId()
+
"
"
+
place.getTokens().size()); } } }
public
void
testCompositeProcesss()
throws
Exception {
long
start
=
System.currentTimeMillis();
for
(
int
i
=
0
; i
<
total; i
++
) { Token token1
=
wm.startWorkFlow(
"
composite
"
); token1.setAttribute(
"
name
"
,
"
dennis
"
); token1.setAttribute(
"
num
"
,
21
); wm.doAction(token1.getId()); assertTrue(token1.isFinished()); }
long
end
=
System.currentTimeMillis(); System.out.println(
"
创建
"
+
total
+
"
个流程运行完毕\n花费时间:
"
+
(end
-
start)
/
1000.0
+
"
秒
"
); }
class
FlowThread
extends
Thread { @Override
public
void
run() {
try
{ barrier.await();
//
wm = new BasicWorkflowManager();
Token token1
=
wm.startWorkFlow(
"
composite
"
); token1.setAttribute(
"
name
"
,
"
dennis
"
); token1.setAttribute(
"
num
"
,
21
); wm.doAction(token1.getId()); assertTrue(token1.isFinished()); barrier.await(); }
catch
(Exception e) {
throw
new
RuntimeException(e); } } } }
文章转自庄周梦蝶 ，原文发布时间2007-10-12
