[one-users] Transfer manager

[Javier Fontan]

Hello,

We already have to finish the documentation but here are some tips.  
The best way to do that is to create your own transfer scripts so  
instead of cloning it just creates a symbolic link in the correct  
directory (VM_DIR/images/disk.<number>). You can inspect scripts in  
lib/tm_commands/nfs, the one that is interesting for you is  
tm_clone.sh. There is also another way to do that and is telling ONE  
to not clone that specific image. I think it is better that you use  
this method until the documentation is finished. A new parameter  
should be added to the disk description (CLONE="no").

--8<------
DISK = [
   source   = "/<some_shared_path>/disk.img",
   target   = "sda1",
   clone    = "no",
   readonly = "no" ]
------>8--

For your information, the new transfer manager is able to create swap  
disk images on the fly, and it is very fast doing that. This example  
creates a 1 Gb swap image and will be mounted in sdb:

--8<------
DISK = [
     type="swap",
     size=1024,
     target="sdb"
]
------>8--

There will be more information on the documentation on all this issues  
but I hope this clarifies the functionality enough to deal with your  
problem. Don't hesitate in contacting us for any doubt you may have.

Bye

On Nov 20, 2008, at 1:12 AM, Alexandre di Costanzo wrote:

> Hi all,
>
> Thank you for all your answers.
>
> I am working with Marcos and we have noticed that you have  
> implemented a disk image transfer mechanism in Open Nebula. Copying  
> by NFS or SSH the disk image each time the user submits a VM.
>
> Within the system that we are currently working on, we have already  
> implemented a similar mechanism to address this need. It is done by  
> our system because we don't want to modify too much Open Nebula,  
> which is already doing a lot things. Hence, we have done some  
> experiments and what we observed is that it takes a very long time  
> to copy the disk image each time (specially with NFS), about 1  
> minute and half for a file of 2GB. It is certainly due to our cheap  
> testbed: 4 linux workstation on a 100Mb network.
>
> In order to hide this lag, we have implemented in our system a cache  
> that pre-copy disk images before submitting a VM, and start a new  
> copy each time a VM is started for having a disk image ready for a  
> the next submission. We can do that because we manage a template  
> directory that the user can choose which template to submit. I will  
> sent more details later if you are interested (I am writing a  
> technical report describing what we are doing).
>
> Thus, our wonder is may you set the use of Transfer manager  
> optional? In other word, let the user decide if he wants to use the  
> TM or not. Because it seams to be mandatory when we add a host to  
> set the TM, or I missed something?
>
> However, it is great a news for us that you handle the disk image  
> transfer.

-- 
Javier Fontan, Grid & Virtualization Technology Engineer/Researcher
DSA Research Group: http://dsa-research.org
Globus GridWay Metascheduler: http://www.GridWay.org
OpenNebula Virtual Infrastructure Engine: http://www.OpenNebula.org