[rescue] Block size and the single DD
jkunz at unixag-kl.fh-kl.de
Wed Feb 11 11:07:15 CST 2004
On Wed, 11 Feb 2004 10:31:11 -0500
"Sheldon T. Hall" <shel at cmhcsys.com> wrote:
> Ah! Maybe I'll turn compression _off_, then. The DLT is far bigger
> than any filesystem I'll be backing up....
This may be a solution. I did this when I had the DLT on my Alpha to
keep it streaming.
> I would assume an Octane is a bit faster than a SPARCclassic or an
It is a 300 MHz R12k machine. It is more then 1.5 times as fast as an
Indigo2 195 MHz R10k and a _lot_ faster then a SPARCclassic with its
extra slow MicroSPARC I.
> Has anyone some suggestions on what "rescue-compatible" hardware might
> be fast enough? My SGI Challenge L with 2xR10k/195 instead of
> 4xR4k/150? An Indigo^2 R10k/195? Sun Ultra 1 170e?
I asume the limiting factor here is not the CPU, but the disk and IO
> Or ... If you're using DLT and keeping it streaming, what have you got
> the DLT connected to?
If you backup local disks try to connect the DLT to an extra SCSI bus.
But mostly you need fast disks. Measure the throughput you get from
A: xfsdump -f - | dd of=/dev/null bs=64k
B: dd if=/dev/rdsk/dks0d1vol of=/dev/null bs=64k count=16384
C: dd if=/dev/rdsk/dks0d1vol of=/dev/tape bs=64k count=16384
B should get you the raw disk speed for linear reads. C should get you
the raw speed of the DLT. If B is about as much as C your disk is way to
slow. If B is more then C, C is the real raw DLT speed. (The DLT should
stream while doing 3) Compare 3 to 1...
More information about the rescue