Ultrium tape throughput on Poweredge

Ben Scott dragonhawk at gmail.com
Fri Jan 5 16:29:15 CST 2007


On 1/5/07, Jose_De_La_Rosa at dell.com <Jose_De_La_Rosa at dell.com> wrote:
> You could use tar to back up directly to tape, and then time it. Use
> something like:
>
> $ time -p tar cvf /dev/nst0 <dir-to-backup>

  The above is a valid and useful test, but it can also be useful to
benchmark the throughput of the device and its I/O path alone (the
above may be limited by the filesystem, etc.).  The following will
write 1 GiB of zeros to the tape, in 64 KiB blocks:

	time dd if=/dev/zero of=/dev/nst0 bs=64k count=16384

  Turning drive hardware compression on and off with the above test
can be useful.  Blocks of zeros tend to compress well, so with
compression on, the tape drive may appear to perform better than it
will with real data.  On the other hand, that would then test how well
the I/O path to the drive can perform (i.e., how fast can the host
feed zeros to the drive).

  Also useful is to test by reading raw blocks from disk and writing
them to tape.  That can test throughput from device to device, without
involving the filesystem or file organization.  The following will
write 1 GiB worth of disk data to the tape, in 64 KiB blocks:

	time dd if=/dev/sda of=/dev/nst0 bs=64k count=16384

  Use caution with the above command.  If you accidentally tell dd to
use the disk as the output file (instead of the input file), it will
happily overwrite your disk with garbage.

  Comparing the performance between zeros, raw blocks from disk, and
tar can help tell you if the performance limits are in the tape
subsystem, the disk subsystem, or filesystem.

-- Ben



More information about the Linux-PowerEdge mailing list