Hi Stefan,
Ok, let me briefly summarize the background of my problem.
I've run into some inconsistency in write times between c/c++  and julia 
version of my toy test. The difference is only apparent when I'm saving a 
file to a network drive. The timings for the local write on the other hand 
for c++ and julia version are more or less the same. I don't have numbers 
in front of me but recall the network write of a ~100MB file was 3-5 times 
slower (need to verify that again today)
I'm total noob in the network field, but could it be a network issue? If so 
I don't get why c++ performance does not drop as well? Is there some sort 
of  caching coming into play....?
Since I am using fwrite() function in C++ version of my test I though it 
would be fair to see how this performs in Julia. But then I run into 
problems with ccall... :)

In the end of the day this is not such a big deal for me and I don't want 
to take up too much of your time, since the network setup I'm in will be 
difficult/impossible to reproduce.
But if anyone is interested to see if the performance is more consistent on 
his end I will post two versions later on today.

Thank you,
kuba

On Friday, February 20, 2015 at 9:32:23 AM UTC-8, Stefan Karpinski wrote:
>
> Also, why are you trying to do your own I/O instead of just using Julia's 
> built-in I/O functionality? There's not a lot of situations where you need 
> to do that.
>
> On Fri, Feb 20, 2015 at 12:30 PM, Stefan Karpinski <[email protected] 
> <javascript:>> wrote:
>
>> On Fri, Feb 20, 2015 at 3:04 AM, Kuba Roth <[email protected] 
>> <javascript:>> wrote:
>>
>>> Ah indeed I got a crash. Thanks for clarification. I've been following 
>>> ccall examples in iostream.jl. Now it makes sense why the ios_write 
>>> function is used there. 
>>> I was wondering does Julia's stream related C functions (ios_write/read) 
>>> have any overhead over their standard C counterparts? 
>>>
>>
>> Certainly not compared to actually doing I/O. It's an alternative to 
>> fwrite which can be as fast or faster – both ultimately make system calls 
>> that actually send data to the kernel. It's pretty common for C libraries 
>> to have their own alternatives to fwrite, which are kind of limited and 
>> basic.
>>
>
>

Reply via email to