[one-users] SCP bottleneck
stefan.freitag at udo.edu
Tue Sep 21 03:57:38 PDT 2010
in Dortmund we have ~250 physical nodes and transfer at the moment the VM
(8 GByte per Image) images via SSH to the nodes. Due to the load caused by
the SSH transfers we limit the number of VMs deployed in parallel to two.
Three or four VMs in parallel should also be possible, but (hopefully) in
very near future we will migrate to another setup:
then scp-wave will be used in the background to transfer the images to the
physical nodes and a "home-brewn" tm mechanism will put in place to start
the VMs on the nodes.
Luis M. Carril schrieb:
> We've a OpenNebula 2.0 cluster in SSH mode and when we deploy near
> simultaneously 3 or 4 MVs of about 1GB, sometimes some of the
> deployments fail because SSH/SCP connection timeouts.
> I've seen in diferent slides that is possible to make TM plugins
> based on parallel-ssh or bittorrent.
> Has anyone experimented similar problems, or developed some of
> these plugins?
> Best regards
> Luis M. Carril
> Técnico Superior
> Centro de Supercomputacion de Galicia (CESGA)
> Avda. de Vigo s/n
> 15706 Santiago de Compostela
> Tel: 34-981569810
> lmcarril at cesga.es
> Users mailing list
> Users at lists.opennebula.org
Dipl.-Inform. Stefan Freitag
Institut für Roboterforschung - Abteilung Informationstechnik
Universitaet Dortmund | phone: +49 (0)231-755-7009
Otto-Hahn-Str. 8 | mobile: +49 (0)176-23900074
44221 Dortmund, Germany | fax: +49 (0)231-755-3251
www.irf.uni-dortmund.de | email: stefan.freitag at udo.edu
More information about the Users