AutoHotkey Community

[qtmspin]

I've run into a limitation and annoying 'bug' It seems if I send messages too fast, ie in a loop like the one below... only the first two get to the client, the other 3 are lost in cyberspace...

If I put a pause of 200 ms in the server code all 5 are delivered to the client. I've tried changing the "ReceivedDataSize = 4096" to both higher and smaller numbers, but nothing has worked.

server code:

[Zed Gecko]

Well i´ve had no time-critical tasks for this script right now,
so i´m wildly guessing here.

First i would suppose, that you set the scripts speed at maximum with

[qtmspin]

It's definatetly the receiving part (my client). I did like you said and hooked up telnet, wow that was easy and cool too! The server is sending everything, however the client is only getting the first two lines.

I already used, Setbatchline -1
I've also got the priority set to realtime
I tried using a loop like this:

[PhiLho]

Well, you use a scripting language, it is not as fast as a native C program, so you must live with its limitations (speed, memory use, etc.).
_________________
vPhiLho := RegExReplace("Philippe Lhoste", "^(\w{3})\w*\s+\b(\w{3})\w*$", "$1$2")

[qtmspin]

I've noticed that when I hook telnet up to the server I get different data then what seems to be delivered to my client.

I tested it out w/ Zed's original script and I get the same thing. Here is what happens...

If I send the text 123456 from the server, 123456 arrives at the client (telnet terminal)

However if I send a shorter string next ie (1234)

123456 arrives at the client (telnet)

I am wondering if the 'pipe' is not being emptied, or perhaps this is a telnet buffer issue?

[Zed Gecko]

Well,
you have probably reached the limits of this language here.
As i can see from the code you have send me,
your ReceiveData function is quite long, and even includes
some MsgBoxes.
So i suppose the buffer of the socket gets filled up while your function
is running.
It would maybe help to make the ReceiveData function just add the received data to a var
and let another (timed) subroutine work with the
data from this var.
But this could produce a lot of new problems, and i´m not shure, if it
would really reduce the "round trip time".

There might be a way to increase the buffer of the socket with the
DLLCall of setsockopt with the value SO_RCVBUF:
http://msdn2.microsoft.com/en-us/library/ms740476.aspx
but i don´t know how to handle it.

The MSDN gives some general hints on speeding up network applications:
http://msdn2.microsoft.com/en-us/library/ms738559.aspx

But the safest way would be probably multithreading, which is not really
possible in AHK (we would need something like #MaxThreadsPerHotkey for OnMessage functions).

[wtg]

[wtg]

Ok, I'm probably showing my ignorance here but I'll ask some questions that might at least serve as a learning tool for those following along. I"m basing my comments on the sample server and client. I have no idea how qtmspin may have altered the programs for his work. And as I said above, it's literally been years since I did socket programming and then it was on unix, not Windows.

Asynch sockets are complicated. Doesn't the server need to handle on FD_WRITE event so that it knows when it's ok for it to write to the socket? By using WSAAsyncSelect you've ensured that the send command won't block when a buffer is full or anything else that would otherwise make it wait, so the FD_WRITE message handler should be used so that we know it's ok to send.

Again, I'm far from an expert but I think you need a few other message handlers setup and need some error handling after the send command. Right now no checking is being done to know whether the send was successful or not.

I did a quick search and found this little primer on basic asynch socket programming, and includes a better explanation of the FD_WRITE event handling. http://www.gamedev.net/reference/articles/article1297.asp

[qtmspin]

I have two sub-routines work with the data now. If the subroutines are working with the data while more data is received, will the received data be ignored? Or are the surroutines treated seperatetly (different thread?) then the Received Data function?

[Chris]

You're probably already aware of the Critical command. It will buffer incoming messages to prevent them from being dropped when a subroutine takes too long to finish. The remarks on the OnMessage page give some details about this.

[wtg]

I did some testing on my own. I learned a little bit, but I'm not sure what.

I modified the server to send the text in a loop i number of times, and to do so with a delay of j ms, implemented with sleep, j. I modified the gui so that I could specify both i and j. I tested with client and server running on the same computer.

The send code is very simple and is shown below. I did not implement an FD_WRITE event, hoping to prove send() would fail if called in a loop too quickly. I thought for sure I'd receive a WSAEWOULDBLOCK error if the loop was too quick.

[wtg]

I found another nice site for info on Winsock programming: http://tangentsoft.net/wskfaq/

[wtg]

So I did some more testing, adding a counter to the messages I sent so that I could see which ones were being lost. I wish I could explain this.



I'd love to know what is going on.

[qtmspin]

Great work, I love it! Have you tried hooking telnet up to the server, its easy and gives great feedback - although I am not sure if it's 100% accurate...

Im sure you know this already but...
start - run - telnet
in telnet type in open 192.168.0.x and the port of the server

[wtg]