[rescue] Block size and the single DD
Sheldon T. Hall
shel at cmhcsys.com
Wed Feb 11 11:27:59 CST 2004
Joshua Boyd said ...
> Looking back on the command you are trying to use:
> > rsh $MACHINE xfsdump -l 0 -F - $FS | dd of=$TAPE bs=$BS
> Why don't you try doing something like:
> rsh $MACHINE xfsdump -l 0 -F - $FS > /dev/null, and see what the
> performance is like that way.
> Also, is xfsdump fast enough? Maybe just on the Challenge do something
> like xfsdump -lo -F - $FS > /dev/null and see if maybe xfsdump isn't
> generating the data fast enough.
I'll do that, too.
Then I'll compare the numbers.
Although I've done some testing on this, I didn't consider using /dev/null
as the "target device". Doing so would certainly save wear-and-tear on the
> Assuming that the network is fine, then I wonder if perhaps switching
> from bs=whatever to ibs=something and obs=somethingelse would be
> sensible. I say this because I'm wondering if dd is blocking somehow in
> appropriately. Perhaps you should have ibs be something really small,
> and obs be something a few K in size? There might be a way to look up
> what the idea outgoing byte size is for the DLT, and I would imagine
> the ideal incoming byte size is going to be how much rsh can transfer
> per packet, perhaps something less than 1024b, like maybe 512b? In my
> experience, matchings block sizes of devices can make dd perform much,
> much faster.
I'll do some research on that, too.
> >From a different post:
> > The SPARCclassic _is_ the remote backup server, as viewed from the SGIs,
> > albeit a remote backup server without a whole lot of disk space. Or CPU
> > power. Or network bandwidth. It is, however, what I have. It also
> > runs 24/7, which makes running "pull" backups from cron pretty easy.
> I think the ideal would be to find a way to cache things on disk. I
> realize that you don't want to spend any money, but putting another 9
> gig or 18gig disk on the SS Classic so that you can copy the filesystem
> to local disk then go from local disk to the DLT. It will obviously
> cost a few dollars, but it will break the problem down into smaller
It's not that I don't _want_ to spend money ....
However, I do have enough space to try that idea, once I tune the Classic so
it will stream the tape from local disk!
> But then, take all of the above with the grain of salt that I haven't
> tried to use a DLT, but I have spent a lot of time trying to make pipe
> things about work faster.
I like your suggestions, and I'm going to use 'em, salt or not.
More information about the rescue