* Dustin J. Mitchell <dustin AT zmanda DOT com> [20090612 11:47]:
> On Fri, Jun 12, 2009 at 11:27 AM, Jean-Francois
> Malouin<malin AT bic.mni.mcgill DOT ca> wrote:
> > I can't get better than ~60MiBs to LTO4 tape when using amtapetype.
> > I tried 32k 1024k and 2048k, with and without hw compression enabled.
> > Any hints? I've added some info below. I can provide more upon request.
> 
> 60 MiB/s is certainly capable of hitting a number of hardware
> bottlenecks - RAM, SCSI, PCI bus, even CPU when generating random
> data.  You may want to try a simpler test with just 'dd' from e.g,
> /dev/zero and /dev/urandom.

I have a exact clone of that library with same LTO4 drives and I can
routinely achieve ~100MiBs on that server. This server is newer and
faster but the (older) machine I'm using for this test used to have
LTO3s and could get more than 60MiBs for those.

The only difference for the drives are the firmware revisions.

I did try with dd and /dev/zero on a non-compression device and I got
more or less the same results.

I'll hook up the new changer to the server with the clone (the one
that performs to specs) and redo the tests to remove issues related to
the presently used computer, a 3 years old 2xdual core Xeon (Conroe).

thanks for your time!
jf

> 
> > Note that tapeinfo reports hw compression off but amtapetype says it
> > on. The changer web interface also tells me that the drive has
> > hardware compression off too. Puzzling.
> 
> amtapetype detects compression by noting that it can fit more data on
> tape when using a fixed byte pattern than when writing pseudo-random
> data.
> 
> Dustin
> 
> -- 
> Open Source Storage Engineer
> http://www.zmanda.com

-- 
<° >< Jean-François Malouin          McConnell Brain Imaging Centre        
Systems/Network Administrator       Montréal Neurological Institute
3801 Rue University, Suite WB219, Montréal, Québec, H3A 2B4, Canada