This is the mail archive of the mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Sparse file performance (Was: Re: Sparse file criteria malfunction- binutils produces sparse .exe & .dll files)

Christopher Faylor wrote:
I remade the executables in an old version of inetutils.
The numbers below show that only the larger ones are sparse
(so the relative overhead is small) and that stripping them
removes sparseness.

This is exactly the kind of data I was looking for.  It seems to me that
this suggests that there really is no issue that we have to worry about
in this case.

Out of curiousity, does building the executable 1) without debugging options,
and 2) with the -s option, also result in non-sparse files?

I'll test this myself when I get back to a windows computer but that won't
be for some time.

I really appreciate your checking into this, Pierre.

I did some profiling of sparse-files vs. non-sparse files. I found that accessing sparse files is between 5% and 10% slower (on 1.3.22/Win2000).

I created a 3Meg, 6Meg, 10Meg and 40Meg file using cp /dev/zero. I then copied each file using windows explorer (and then verified that the sparse bit was gone).

Then I ran 'time cat filename > /dev/null' (i ran it a few times to make sure the file was cached). The performance difference was:

40Meg: 5%
10Meg: 7%
6Meg: 10%
3Meg:  5%

This wasn't the most sofisticated test ever, I did not ensure that the files were equally fragmented on disk. But, it does show that sparse files are notably slower.

Unsubscribe info:
Problem reports:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]