View mode: basic / threaded / horizontal-split · Log in · Help
June 21, 2012
filename.writeln() across network
I wrote a program that parses a text file and writes results as 
it is processing the file (i.e. many writeln()'s).  On my local 
harddrive it works fine.  When I later used it on a file located 
on a file server, it went from 500ms to 1 minute processing time.

It there a more efficient way to write out the results that would 
say maybe only access the harddrive as it closes the 
connection...or somesuch?

Thanks for your assistance.
June 21, 2012
Re: filename.writeln() across network
On Thu, 21 Jun 2012 14:56:37 +0100, Paul <phshaffer@gmail.com> wrote:

> I wrote a program that parses a text file and writes results as it is  
> processing the file (i.e. many writeln()'s).  On my local harddrive it  
> works fine.  When I later used it on a file located on a file server, it  
> went from 500ms to 1 minute processing time.
>
> It there a more efficient way to write out the results that would say  
> maybe only access the harddrive as it closes the connection...or  
> somesuch?
>
> Thanks for your assistance.

I imagine writeln is synchronous/non-overlapped IO.  Meaning, the call to  
writeln doesn't return until the write has "completed".  So, on every call  
you're basically waiting for the network IO to complete before you process  
something else locally.

What you want is asynchronous or overlapped IO where the write starts, and  
the function returns, and then you later get notified that the write has  
completed.  This lets you continue processing locally while the write  
happens in the background.

That's the theory, in practice I'm not sure what options you have in  
phobos for overlapped IO.  If you're on windows you can pull in the win32  
functions CreateFile, WriteFile etc and define the data structures  
required for overlapped IO.

R

-- 
Using Opera's revolutionary email client: http://www.opera.com/mail/
June 21, 2012
Re: filename.writeln() across network
On Thursday, June 21, 2012 18:14:26 Regan Heath wrote:
> On Thu, 21 Jun 2012 14:56:37 +0100, Paul <phshaffer@gmail.com> wrote:
> > I wrote a program that parses a text file and writes results as it is
> > processing the file (i.e. many writeln()'s). On my local harddrive it
> > works fine. When I later used it on a file located on a file server, it
> > went from 500ms to 1 minute processing time.
> > 
> > It there a more efficient way to write out the results that would say
> > maybe only access the harddrive as it closes the connection...or
> > somesuch?
> > 
> > Thanks for your assistance.
> 
> I imagine writeln is synchronous/non-overlapped IO. Meaning, the call to
> writeln doesn't return until the write has "completed". So, on every call
> you're basically waiting for the network IO to complete before you process
> something else locally.
> 
> What you want is asynchronous or overlapped IO where the write starts, and
> the function returns, and then you later get notified that the write has
> completed. This lets you continue processing locally while the write
> happens in the background.
> 
> That's the theory, in practice I'm not sure what options you have in
> phobos for overlapped IO. If you're on windows you can pull in the win32
> functions CreateFile, WriteFile etc and define the data structures
> required for overlapped IO.

If he's pulling in data from the network and then writing it to disk, he could 
use std.concurrency to have the network stuff on one thread and the writing on 
another. It would probably mean copying the data to be able to pass it across 
threads (unless he's using different buffers every time he's reading the data 
from the network, in which case casting to immutable could do the trick), but 
it would disconnect the reading from the writing. IIRC, there's an example in 
TDPL's concurrency chapter on how to use std.concurrency to read and write 
files concurrently, which could be used as a starting point.

http://www.informit.com/articles/article.aspx?p=1609144

- Jonathan M Davis
June 21, 2012
Re: filename.writeln() across network
On Thursday, 21 June 2012 at 17:14:34 UTC, Regan Heath wrote:
> On Thu, 21 Jun 2012 14:56:37 +0100, Paul <phshaffer@gmail.com> 
> wrote:
>
>> I wrote a program that parses a text file and writes results 
>> as it is processing the file (i.e. many writeln()'s).  On my 
>> local harddrive it works fine.  When I later used it on a file 
>> located on a file server, it went from 500ms to 1 minute 
>> processing time.
>>
>> It there a more efficient way to write out the results that 
>> would say maybe only access the harddrive as it closes the 
>> connection...or somesuch?
>>
>> Thanks for your assistance.
>
> I imagine writeln is synchronous/non-overlapped IO.  Meaning, 
> the call to writeln doesn't return until the write has 
> "completed".  So, on every call you're basically waiting for 
> the network IO to complete before you process something else 
> locally.

Isn't the most simple approach then to build up the whole file in
memory as a single string, using \n and then do a single write
across the network ?

like:

void main(string args[]){
  string filecontent = "";
  filecontent ~= "#include std.stdio\n";
  filecontent ~= "int x = " ~ x ~ ";\n";
  //etc etc...

  auto f = new File("X:\\MyNetworkDir\\file.txt");
  f.writeln(filecontent);
  f.close();
}

Haven't tested the code, but if network IO is the wait, then just
increase the buffer..
June 26, 2012
Re: filename.writeln() across network
On Thursday, 21 June 2012 at 19:52:26 UTC, Danny Arends wrote:
> On Thursday, 21 June 2012 at 17:14:34 UTC, Regan Heath wrote:
>> On Thu, 21 Jun 2012 14:56:37 +0100, Paul <phshaffer@gmail.com> 
>> wrote:
>>
>>> I wrote a program that parses a text file and writes results 
>>> as it is processing the file (i.e. many writeln()'s).  On my 
>>> local harddrive it works fine.  When I later used it on a 
>>> file located on a file server, it went from 500ms to 1 minute 
>>> processing time.
>>>
>>> It there a more efficient way to write out the results that 
>>> would say maybe only access the harddrive as it closes the 
>>> connection...or somesuch?
>>>
>>> Thanks for your assistance.
>>
>> I imagine writeln is synchronous/non-overlapped IO.  Meaning, 
>> the call to writeln doesn't return until the write has 
>> "completed".  So, on every call you're basically waiting for 
>> the network IO to complete before you process something else 
>> locally.
>
> Isn't the most simple approach then to build up the whole file 
> in
> memory as a single string, using \n and then do a single write
> across the network ?
>
> like:
>
> void main(string args[]){
>   string filecontent = "";
>   filecontent ~= "#include std.stdio\n";
>   filecontent ~= "int x = " ~ x ~ ";\n";
>   //etc etc...
>
>   auto f = new File("X:\\MyNetworkDir\\file.txt");
>   f.writeln(filecontent);
>   f.close();
> }
>
> Haven't tested the code, but if network IO is the wait, then 
> just
> increase the buffer..

Thanks for the idea.  I thought maybe there would be a way to use 
writeln() to do what you illustrated...basically writing to a 
buffer and then write the buffer to disk when needed.  It would 
be a nice feature that would allow a developer to change the 
character of his file writing quickly w/o changing much code.  
Maybe open a file File("filemane.txt", "buffer") or 
somesuch...and then a way to tell it to write to disk when you 
want.
Top | Discussion index | About this forum | D home