W3C home > Mailing lists > Public > www-ws-arch@w3.org > October 2002

RE: How to determine if an operation is synchronous

From: Assaf Arkin <arkin@intalio.com>
Date: Tue, 22 Oct 2002 17:47:01 -0700
To: "Ugo Corda" <UCorda@SeeBeyond.com>
Cc: <www-ws-arch@w3.org>
Message-ID: <IGEJLEPAJBPHKACOOKHNGEAKCMAA.arkin@intalio.com>

WSDL 1.1 does not define the perspective of the client, only that of the
server. The document you are referencing does a better job of explaining how
a client would logically behave.

From the perspective of the client, if you are doing a one-way operation (as
defined by the server) then you are putting a message out there that would
be sent and received by the server. The latency between the client actually
sending the message and the server receiving it is an aspect of the protocol
binding. The protocol can have no latency, or it may have some latency. For
example, if you're sending using SMTP, the client can tell when the message
was sent, not when it was received. If you are sending using HTTP you can
tell when it was received.

However, by looking only at the operation definition without the protocol
binding, you can only tell that the operation is asynchronous.

If you are doing a request-response operation (as defined by the server)
then you are putting a message out there and you may or may not wait for the
response. You are guaranteed that if the operation was performed by the
server, then the message was received and a response was generated (that's
what WSDL 1.1 says about request-response). It would be logical of you to
wait for the response to come back, if nothing else, then to know that the
operation was performed. It does not make sense to design a client that does
not wait for the response, and the IBM document is just written given that
assumption.

From a WSDL operation definition you can determine that it is:

- asynchronous  The operation involves a single message (one-way and
notification in WSDL 1.1)
- synchronous  The operation involves two messages in opposite directions,
and indicates that it completes after receiving one and sending the other
(request-response and solicit-response in WSDL 1.1)

arkin



Arkin,
Your definition of synchronous/asynchronous does not seem to be the same as
the commonly used one for Web services operations (see for example [1]).
This commonly used definition is basically about the client blocking or not
after submitting a SOAP request (and such behavior is not described by WSDL,
as far as I can see).

Assuming we adopt your definition, could you please explain how WSDL
provides the information that would allow me to tell a synchronous operation
from an asynchronous one (according to your definition)?

Thank you,
Ugo


[1] http://www-106.ibm.com/developerworks/webservices/library/ws-asynch1/

-----Original Message-----
From: Assaf Arkin [mailto:arkin@intalio.com]
Sent: Tuesday, October 22, 2002 4:20 PM
To: Ugo Corda
Cc: www-ws-arch@w3.org
Subject: How to determine if an operation is synchronous


Let's assume two nodes, A and B, that exchange a message through some
protocol(s) of unknown latency L.


Asynchronous:

A sends a message M1 at time T1, B receives the message M1 at time T2. T2 =
T1+L.

Since neither A nor B know what L is, A cannot determine when B received the
message, nor can B determine when A sent the message.

(The message may contain a time stamp equal to T1, but it's possible that
A's clock is off compared to B's clock, and so the time reading is not
reliable)


Synchronous:

A sends a message M1 at time T1, B receives the message M1 at time T2. Again
T2 = T1+L. B processes the message, the processing time P is unknown as
well. B sends back the message M2 at time T3. T3 = T2+P = T1+L+P. A receives
the message M2 at time T4. T4 = T3+L = T1+2*L+P.

Since A and B do not know L, A cannot determine when B received M1 or sent
M2, nor can B determine when A send M1 or received M2.

A knows that its operation occurred around the time instant Ta. Ta is
calculated from the mean time of the operation. Ta =
(T1+T4)/2.

B knows that its operation occurred around the time instant Tb. Tb is
calculated from the mean time of the operation. Tb = (T2+T3)/2.

Ta = (T1+T4)/2 = (T1+T1+2*L+P)/2 = T1+L+P/2 =(T1+L+T1+L+P)/2 = (T2+T3)/2 =
Tb

So, whatever L and P are, both operations occur at the same time instant. We
say that A and B are synchronized.

Now, if A included a time stamp in the message M1 (T1), and B included two
time stamps in the message M2 (T2, T3), then A could synchronize it's clock
against B. A can't just use the values T2 or T3, because A's clock can be
off compared to B's.

But A knows that it's operation occurred at Ta (it's clock), and knows that
Ta=Tb+offset, and knows that Tb=(T2+T3)/2 (T2,T3 given from B's clock), so
it knows the offset is Ta-(T2+T3)/2. A can now synchronize it's clock with
B's.

arkin
Received on Tuesday, 22 October 2002 20:46:03 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 3 July 2007 12:25:09 GMT