Classic running out of memory... huh? what? <long>

[Tom Buskey]

On Thu, Jun 11, 2009 at 3:25 PM, Lloyd Kvam <python at venix.com> wrote:

> On Thu, 2009-06-11 at 13:41 -0400, bruce.labitt at autoliv.com wrote:
> > I bet it won't do 480Mbits sustained.
> > (For a 10 GB file write) However, I just might try it for the heck of
> > it!
> I use external USB and firewire drives fairly regularly.  USB does not
> come close to achieving 48 MB / sec.  Firewire (400 mb flavor) does
> noticeably better than USB.


That's what I'd expect.


>
> Would dd from /dev/zero to /dev/sdX would be a reasonable way to
> benchmark the interface at the expense of any data on the drive?


I've done that to do a quick test across the network, to tape, etc.  You
need to watch out for cache effects; your data stream needs to be larger
then the cache.

Linux does some things to speed up /dev/zero that others don't (IIRC).
Using dd from /dev/zero to a compressed ZFS file system will be
*significantly* faster then real data ;-)

There are better tests out there.

(I suppose higher quality hardware might do better.  I assume these are
> commodity parts and essentially buy what's on sale.)
>

Plugging in something quick might be ok for a test, but imagine you're
pushing 190 GB there on a regular basis.  I had to restore from a standalone
LTO3 drive to a local SCSI disk.  It took 36 hours.  If I have to do it on a
regular basis, I don't want to wait that long.

The specs on the tape drive/scsi controller say 20 MB/s so it should take..

190 GB * (1 GB/1024 MB)  * (1 sec / 20 MB) * (1 min/60 sec) * (1 hr/60 min)
=

2.72 hours.

I think I got closer to 1.5 MB/s
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.gnhlug.org/mailman/private/gnhlug-discuss/attachments/20090611/d3152ee2/attachment.html