export disks from nodes?

Sean Dilda agrajag at scyld.com
Wed Jan 9 10:50:55 PST 2002


On Wed, 09 Jan 2002, hanzl at noel.feld.cvut.cz wrote:

> My application is rather special cause it is nearly perfect regarding
> HD/CPU communication - it is quite happy with every HD streaming data
> to CPU in the same box for tens of minutes, then results are quickly
> mixed using network and all this repeats. There is not much left for
> PVFS to help with - application itself keeps all HDs busy and all CPUs
> busy, so it makes no sense to send data over network.
> 
> Network access to all data is infrequent, I just want it to be
> possible and relatively easy (easier than to bpcp them).

bpcp does have a recursive mode, so if its a matter of recursively
copying, you can just do that.  Otherwise, if you want easy acces, I
might also suggest having a central store for the data on the master
node that you can easily tweak, and have a script that'll send the data
out to the slave nodes whenever you change anything.  However that'll
require you to copy everything over even for a minor change.   If you do
have a lot of data, you can look into using rsync (with bpsh as its rsh
implementation) instead of bpcp.  rsync will only transfer over what's
changed, thus saving a lot of bandwidth.  I've never tried rsync with
bpsh, but I think it should work.


Sean
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 232 bytes
Desc: not available
Url : http://www.scyld.com/pipermail/beowulf/attachments/20020109/9b94a86e/attachment.bin


More information about the Beowulf mailing list