|
From: Szakacsits S. <sz...@si...> - 2003-04-14 11:56:53
|
On Fri, 11 Apr 2003, Szakacsits Szabolcs wrote: > On Thu, 10 Apr 2003, Ian Jackson wrote: > > > > This can be done, of course, just by piping the output of the imager > > into the restorer. The few resulting extra memory copies will > > probably pale into insigificance next to the overhead of disk reads > > and writes. Although, I'll try it and see. > > I think you're right. If I guess correctly how you implement them, yours > will be even faster. The reason, ntfsclone is suboptimal for disk IO, it > will seek a lot but not yours. It's minutes to fix this and ntfsclone > should be faster. We will see then by how much ... In the weekend I made ntfsclone faster. For me it gave 32% speedup. However the speedup can be dramatically more (or less but must be faster than now) depending on the filesystem fragmentation. Also of course input/output devices and their characteristics, MAGIC hardcoded kernel values, driver and overall the hardware matters but as I've seen the disk IO was pretty saturated (Mandrake 9.1 default kernel) and maximum the CPU usage could be reduced, probably only a bit. There are several ways to improve it further on but I don't see it a high priority especially the benefit is dubious and there are other requested real funs to do. I'll send a cummulative patch to Anton when I also fixed the '-O device' case (this case indeed doesn't work but won't make damage, so restore is possible only through the inefficient stdout) and some other minor issues unless one is interested in the speedup now. Szaka |
