Copying a large file system (again)

Tino Schwarze linux-poweredge.lists at tisc.de
Tue Dec 4 09:06:28 CST 2007


On Tue, Dec 04, 2007 at 08:39:32AM -0500, Kuba Ober wrote:

> > > Thanks, it looks like I'll need to look for some more memory for the
> > > source machine. Maybe I'll try plugging the SATA RAID into the
> > > destination machine directly (a PE1800 with more RAM) and copy locally,
> > > skipping the network, then maybe rsync will be an option since most of
> > > the files are already in place, just the main pool is still missing.
> >
> > Does your destination machine have sufficient memory?  I'm not
> > hard-stuck on rsync, but one of the niceties of running rsync over SSH
> > is that you can do it from either end.
> 
> If you have two PowerEdges nearby and in your control, there's no reason not 
> to use a dedicated gigabit link between the two, without any encryption. SSH 
> is dog-slow and takes up CPU unnecessarily when all you do is move data 
> between adjacent machines (if that's what you do).

I'm using netcat, there's nothing with less overhead. Apart from that,
the source machine's I/O is the bottleneck. It's a lot of little files,
I'm copying the pool with GNU tar now and vmstat often says only 360
blocks read in. It's three WD 500 GB SATA-II disks (Raid Edition) in an
SW-RAID5. The SW-RAID seems to max out at around 25000 or 28000 blocks
read per second. I'm not sure how much a block is though.

> > Failing that, GNU TAR may well work better on less RAM.  S-TAR is
> > nice, but it *might* have a memory leak or something that causes it to
> > utilize so much memory.  
> 
> Again, Joerg might have an answer to that. It may be simply that you didn't 
> give an option of some sort or another :)

I guess, managing several millions of files just takes some memory to
keep track of the hardlinks. GNU tar is at 250 MB now, having copied
half of the pool, but this cannot be compared with star since star started
with another portion of the file system (the individual server's backup
directories).

Bye,

Tino.

-- 
www.craniosacralzentrum.de
www.lebensraum11.de
www.spiritualdesign-chemnitz.de

Tino Schwarze * Parkstraße 17h * 09120 Chemnitz



More information about the Linux-PowerEdge mailing list