Re: [twsocket] [ICS-SSL] built-in timer

2006-07-16 Thread Arno Garrels
Francois PIETTE wrote:
 I would also vote for a built-in timer since any server needs to
 perform some tasks periodically, so probably deriving from
 TComponent plus a window receiving timer messages is a good choice.
 
 Most communication application and many other needs to perform some
 task periodically as well as handle all kind of time out. 

Concerning a client idle timeout in a server application/component 
where would you refresh client's last alive time stamp? It's clear to
refresh it in OnDataAvailable, but that is not enough, it has to be
refreshed also when data is being sent. Should I use/override TryToSend
for this purpose? What if someone would put 2GB into the send buffer? 

 It wouyld
 be usefull to add a general purpose timer component able to do a lot
 of tasks, centralized in a single component. 

I see some problems with V6, since TWndControl does not handle WM_TIMER
messages by default. Such timer should not create a window but attach
to an existing window.

I currently derive a server from TComponent and create a window that is
used for WM_TIMER and other component messages (makes it easy to
ThreadAttach/ThreadDetach the component/timer). If you want to preserve
an option to run non-WSocket-events, and listening as well as client
sockets all in different threads such design appears to be OK.  

 A single TIcsTimer
 component is enough in an application if correctly designed. All
 other components needing a timer could be linked to this TIcsTimer,
 much like many data-aware components are linked to a single
 datasource component or many SQL-Query type component are linked to a
 single database connection component.

Hmm, I see multi-threading issues, since a timer always fires in 
context of the thread where it has been created. 
 
 
 This discussion should probably be moved to the twsocket mailing
 list. Isn't it ?


---
Arno Garrels [TeamICS]
http://www.overbyte.be/eng/overbyte/teamics.html



-- 
To unsubscribe or change your settings for TWSocket mailing list
please goto http://www.elists.org/mailman/listinfo/twsocket
Visit our website at http://www.overbyte.be


Re: [twsocket] [ICS-SSL] built-in timer

2006-07-16 Thread Arno Garrels
Francois PIETTE wrote:
 Francois PIETTE wrote:
 Concerning a client idle timeout in a server application/component
 where would you refresh client's last alive time stamp? It's clear to
 refresh it in OnDataAvailable, but that is not enough, it has to be
 refreshed also when data is being sent. Should I use/override
 TryToSend for this purpose? What if someone would put 2GB into the
 send buffer? 
 
 There is an event triggered when TWSocket send each data pacet:
 OnSendData which is perfect in that case.

I forgot this event completely :) I hope it's not too slow calling 
GetTickCount from it.

 Hmm, I see multi-threading issues, since a timer always fires in
 context of the thread where it has been created.
 
 What if TIcsTimer post a message to the calling component when the
 programmed timer event has occured ?  Wouldn't this would solve all
 multithread issues ?

Then TIcsTimer would have to manage a list of linked ICS-Components.
When the timer fires it has to iterate thru this list. But 
deferencing a component pointer which might have been already destroyed 
(by another thread) will crash, right? You probably can make that 
threadsafe, however I fear it would slow down performance too much?  


---
Arno Garrels [TeamICS]
http://www.overbyte.be/eng/overbyte/teamics.html
   
-- 
To unsubscribe or change your settings for TWSocket mailing list
please goto http://www.elists.org/mailman/listinfo/twsocket
Visit our website at http://www.overbyte.be


Re: [twsocket] built-in timer

2006-07-16 Thread Francois PIETTE
 Hmm, I see multi-threading issues, since a timer always fires in
 context of the thread where it has been created.

 What if TIcsTimer post a message to the calling component when the
 programmed timer event has occured ?  Wouldn't this would solve all
 multithread issues ?

 Then TIcsTimer would have to manage a list of linked ICS-Components.
 When the timer fires it has to iterate thru this list.

There are surely ways to design it to avoid scanning a list of components. A 
standard TTimer is probably not the best design either. Probably a dedicated 
thread could do the job, using MsgWaitForMultipleObjectsEx to make the 
thread sleep until the next event to be signaled. The event to be signaled 
can be the ID in the component list. Or something like that: The idea just 
pops out of my head.

 But deferencing a component pointer which might have been already 
 destroyed
 (by another thread) will crash, right?

Right. Maybe could we impose that all linked component derive from 
TIcsWndControl and implement the required behaviour in TIcsWndControl. That 
is unregister the timeout from the detructor.

 You probably can make that
 threadsafe, however I fear it would slow down performance too much?

I think not much more than using a TTimer which anyway post a message. We 
could also design the thing like MS do: either using a message or using a 
callback (which is called in the other thread context, that is the user is 
responsible for managing the thread concurrency).


--
Contribute to the SSL Effort. Visit http://www.overbyte.be/eng/ssl.html
--
[EMAIL PROTECTED]
http://www.overbyte.be


-- 
To unsubscribe or change your settings for TWSocket mailing list
please goto http://www.elists.org/mailman/listinfo/twsocket
Visit our website at http://www.overbyte.be


Re: [twsocket] UDP Concept doubt

2006-07-16 Thread Francois PIETTE
 Now I got it

UDP packets are not fragmented at the application level, but you may get 
duplicates, or miss some packet, recieve the packets in incorrect order. 
Those conditions are very difficult to reproduce on a small LAN. You have to 
have a large LAN with different speed in different subnet and have several 
routes between points A and B. You are likely to get the problems on the 
internet when communicating between systems with long routes between them 
(Use tracert to know).

It is much easier to deal with TCP fargmentation than to deal with the 
unreliable (by design !) UDP protocol. The only interesting features are 
broadcast and multicast capabilities UDP has an TCP lacks.

Contribute to the SSL Effort. Visit http://www.overbyte.be/eng/ssl.html
--
[EMAIL PROTECTED]
http://www.overbyte.be



- Original Message - 
From: Éric Fleming Bonilha [EMAIL PROTECTED]
To: ICS support mailing twsocket@elists.org
Sent: Tuesday, July 04, 2006 1:44 PM
Subject: Re: [twsocket] UDP Concept doubt


 Thanks Dan!!!

 Now I got it

 Thanks!!!

 - Original Message - 
 From: Dan [EMAIL PROTECTED]
 To: 'ICS support mailing' twsocket@elists.org
 Sent: Monday, July 03, 2006 7:11 PM
 Subject: Re: [twsocket] UDP Concept doubt


 UDP packets can be fragmented at the IP level, but once you do a receive 
 you
 will receive the full datagram so you don't have to worry about it.  A
 problem you may run into though is that of WSockets default buffer not 
 being
 big enough for a full size UDP packet... you might have to set the buffer
 higher than the default if you aren't receiving the full datagrams.

 Dan

 -Original Message-
 From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On
 Behalf Of Éric Fleming Bonilha
 Sent: 03 July 2006 13:42
 To: twsocket@elists.org
 Subject: [twsocket] UDP Concept doubt

 Hello Everyone!

 I´m in doubt in some aspects of UDP.

 Do UDP suffers from fragmentation like TCP? what I mean is that in TCP if 
 I
 send a chunck of data, it is not guaranteed that this data will be 
 received
 in a single event correct? It is guaranteed that it will be delivered
 completed and in order, so we have to implement packet boundaries on the
 application protocol.

 I would like to know if I send packets of data in UDP it will be 
 fragmented
 too, because if it is, than an application protocol that will use UDP as
 underling layer will have to implement some packet boundaries mechanism.

 I´m asking this because of the RTP (Real Time Protocol) protocol, I´m
 currently developing an application that will have to parse RTP packets, 
 but
 on RTP specification there is no packet boundaries. On the specification 
 is
 only said that one RTP packet should be sent on one UDP packet, but I´m
 thinking if UDP packet will be fragmented and will be received in one or
 more events on ICS

 Thanks a lot

 Éric Fleming Bonilha
 -- 
 To unsubscribe or change your settings for TWSocket mailing list
 please goto http://www.elists.org/mailman/listinfo/twsocket
 Visit our website at http://www.overbyte.be


 -- 
 To unsubscribe or change your settings for TWSocket mailing list
 please goto http://www.elists.org/mailman/listinfo/twsocket
 Visit our website at http://www.overbyte.be

 -- 
 To unsubscribe or change your settings for TWSocket mailing list
 please goto http://www.elists.org/mailman/listinfo/twsocket
 Visit our website at http://www.overbyte.be 

-- 
To unsubscribe or change your settings for TWSocket mailing list
please goto http://www.elists.org/mailman/listinfo/twsocket
Visit our website at http://www.overbyte.be