1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<html>
<head>
<title>Asynchronous Method Invocation - CORBA for impatient clients</title>
<!-- -->
</head>
<BODY text = "#000000"
link="#000fff"
vlink="#ff0f0f"
bgcolor="#ffffff">
<h3>Asynchronous Method Invocation - CORBA for impatient clients</h3>
<P>Our <A HREF="../Simple/Client/index.html">simple client</A>
illustrated how to use the traditional CORBA synchronous method
invocation to query a number of stock prices.
Suppose that we had to query hundreds of stock prices, for
example, during the initialization of a complex market analysis
tool.
In that case sending the requests in sequence is going to yield
poor performance; we are not taking advantage of the natural
parallelism in distributed systems, since we are waiting for the
first response to come back before sending the next query.
Traditionally this problem has been attacked using either
<CODE>oneway</CODE> calls or multiple threads. Both approaches can
work, but they have some disadvantages:
multi-threading programming can be hard and error-prone,
oneways are unreliable and require callback interfaces to return
the stock value.
Recently the OMG approved the CORBA Messaging specification
that extends the basic invocation model to include asynchronous
calls.
Unlike the old deferred synchronous model, the new model uses
the IDL compiler and the SII to achieve type safety and improved
performance, but the application does not block waiting for a
response. Instead, it gives the ORB a reference to a reply
handler that will receive the response asynchronously.
The specification also defines a polling interface that we will
not discuss as TAO does not implement it.
</P>
<P>For this problem we will extend the IDL interface to include a
new operation:
</P>
<PRE>
interface Single_Query_Stock : Stock {
double get_price_and_names (out string symbol,
out string full_name);
};
</PRE>
<P>This will simplify some of the examples.
</P>
<P>The first step is to generate the stubs and skeletons with
support for the callback AMI. We do this with the <CODE>-GC</CODE>
flag:
</P>
<PRE>
$ $ACE_ROOT/TAO/TAO_IDL/tao_idl -GC Quoter.idl
</PRE>
<P>You may want to take a brief look at the generated client side
interface.
The IDL compiler adds new methods to the <CODE>Quoter::Stock</CODE>
interface. In particular pay attention to this operation:
</P>
<PRE>
virtual void sendc_get_price_and_names (
AMI_Single_Query_StockHandler_ptr ami_handler
);
</PRE>
<P>This is the operation used to send a request asynchronously. The
response is received in the handler object. This is a regular
CORBA object with the following IDL interface:
</P>
<PRE>
interface AMI_Single_Query_StockHandler {
void get_price_and_names (in double ami_return_val,
in string symbol,
in string full_name);
};
</PRE>
<P>You don't have to write this IDL interface. The IDL compiler
automatically generates the so-called <EM>implied IDL</EM>
constructs from your original IDL.
Notice how the arguments are generated. The first argument is
simply the return value, then the output arguments show up, but
as <EM>input</EM> only since the handler has to receive the
reply.
</P>
<H3>Implementing the reply handler</H3>
<P>We will have to implement a servant for this new IDL interface
so we can receive the reply,
exactly as we do for servers:
</P>
<PRE>
class Single_Query_Stock_Handler_i : public POA_Quoter::AMI_Single_Query_StockHandler
{
public:
Single_Query_Stock_Handler_i (int *response_count)
: response_count_ (response_count) {}
void get_price_and_names (CORBA::Double ami_return_val,
const char *symbol,
const char *full_name)
{
std::cout << "The price of one stock in \""
<< full_name << "\" (" << symbol << ") is "
<< ami_return_val << std::endl;
*this->response_count_++;
}
private:
int *response_count_;
};
</PRE>
<P>The <CODE>response_count_</CODE> field will be used to
terminate the client when all the responses are received.
</P>
<H3>Sending asynchronous method invocations</H3>
<P>The handler servant is activated as any other CORBA object:
</P>
<PRE>
int response_count = 0;
Single_Query_Stock_Handler_i handler_i (&response_count);
Quoter::AMI_Single_Query_StockHandler_var handler =
handler_i._this ();
</PRE>
<P>and now we change the loop to send all the requests at once:
</P>
<PRE>
int request_count = 0;
for (int i = 2; i != argc; ++i) {
try {
// Get the stock object
Quoter::Stock_var tmp =
factory->get_stock (argv[i]);
Quoter::Single_Query_Stock_var stock =
Quoter::Single_Query_Stock::_narrow (tmp.in ());
stock->sendc_get_price_and_names (handler.in ());
request_count++;
}
</PRE>
<P>after the loop we wait until all the responses have arrived:
</P>
<PRE>
while (response_count < request_count
&& orb->work_pending ()) {
orb->perform_work ();
}
</PRE>
<H3>Exercise 1</H3>
<P>Complete the <CODE>client.cpp</CODE> file. Does this client play
the server role too? If not, what is the role with respect to
the handler servant? If you think it is a server too, what
should you do about the POA?
</P>
<P>You can use the following files to complete your implementation:
the <A HREF="Quoter.idl">Quoter.idl</A>,
<A HREF="Handler_i.h">Handler_i.h</A>,
<A HREF="Handler_i.cpp">Handler_i.cpp</A>.
Remember that the simple client main program
(located
<A HREF="../Simple/Client/client.cpp">here</A>)
is a good start.
</P>
<H4>Solution</H4>
<P>Look at
<A HREF="client.cpp">client.cpp</A> file. It should
not be much different from yours.
</P>
<H3>Testing</H3>
<P>A simple server is provided, based on the
<A HREF="../Simple/Server/index.html">simple server</A>
from the introduction.
As before, you need the following files:
<A HREF="Stock_i.h">Stock_i.h</A>,
<A HREF="Stock_i.cpp">Stock_i.cpp</A>,
<A HREF="Stock_Factory_i.h">Stock_Factory_i.h</A>
<A HREF="Stock_Factory_i.cpp">Stock_Factory_i.cpp</A>
and <A HREF="server.cpp">server.cpp</A>.
</P>
<H2>Configuration</H2>
<P>So far we have used the default configuration in TAO,
but AMI works better with some fine tuning. For example,
by default TAO uses a separate connection for each outstanding
request. With SMI this is a very good strategy, as separate
threads can send concurrent requests without any shared
resources,
but this approach does not scale well with AMI, as it would
create too many connections.
The solution is to change the strategy to share connections.
All we need to do is create a
<A HREF="svc.conf">svc.conf</A> file with the following
contents:
</P>
<PRE>
static Client_Strategy_Factory "-ORBTransportMuxStrategy MUXED"
</PRE>
<P>There are many other configuration options, all of them
documented in
<A HREF="../../../Options.html">Options.html</A>,
in
<A HREF="../../../configurations.html">configurations.html</A>,
and in the Developer's Guide available from
<A HREF="http://www.theaceorb.com/">OCI</A>.
</P>
<hr>
<address><a href="mailto:coryan@cs.wustl.edu">Carlos O'Ryan</a></address>
</body>
</html>
|
