From: MikkelFJ <mikkelfj-anti-spam@...>
Date: 2002-11-19T09:48:02+09:00
Subject: Re: Not really a ruby question, but this is a smart group.  Win32 file.write timing.


"MetalOne" <jcb@iteris.com> wrote in message
news:92c59a2c.0211171917.8e4e118@posting.google.com...
> I forgot.
> But when I looked at the C++ code, the write buffer was stored on the
stack.
> Surely that is in physical ram?

I was perhaps assuming you were stressing the system more than you are.

Physical RAM is btw. only part of the deal. Physical RAM outside the
processor cache may also be an issue - although I don't know that much about
how the filesystem driver might bypass processor cache using DMA.

My general unscientific impression is that Windows NT, NTFS is rather fast
at dumping data to a file if the conditions right such as perfectly cached
data and no fragmentation problems etc. But then you are concerned with the
worst case to which I can only say that I doubt it takes 150ms to write 150K
perfectly cached data in a system that is in no way stressed.

BTW: NT writes in 64K blocks whether you have modified 1 byte or 64K byte of
the block.
There may also be issues with flushing. It could be that you are measuring a
number of very fast writes and then suddenly find yourself measring the time
where flushing happens for much more data than you expect.


Mikkel