"[email protected]" <[email protected]> wrote:

> >> Now that is a 1 to 5 performance gap !!
> >> Any ideas about how to explain this ?
> >>     
> >
> > Are you sure that linux has actually written the data to disk
> > when the dd has returned?  In most cases (esp with such
> > a small amount of data) linux has the data in it's buffer
> > cache and *not* on the disk.
> >   
> dd on linux with /dev/zero does a little trick where they don't actually do
> any copies.  Try adding "conv=ascii" to the dd command.

If this was true, then dd in Linux would not be usable because it does not what
dd is expectec to do.

>From my experiences, filesystems on Linux are slow and Linux tries to create 
the impression that is is fast. Many people unpack the Linux kernel tar archive
and call this a "benchmark". If you carefully check disk access during the time 
when gnu tar is "working", you see that the writes to disk start abter gnu tar 
finished, so the time for this tar run is completely meanlingless. If you use 
star for this task, you notice that star (by default) runs 4x longer because
star calls fsync(2) at the end of each extracted file. If you call 

star -x -no-fsync

it runs at least at the speed of gnu tar.

It may be a good idea to check the disk access and to call sync(1) after the dd 
test.....

Jörg

-- 
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected] (work) Blog: 
http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/private/ ftp://ftp.berlios.de/pub/schily
_______________________________________________
opensolaris-discuss mailing list
[email protected]

Reply via email to