So here's the input for the discussion re RHUL upgrade.
Greig wrote:
---
> I would like to understand why we need to migrate the classicSE
> data into DPM at all. If we keep the nfs mount people can still
> access their files using their lcg-* gfal commands since the
> gridftp interface is still there and the physical path does not change.
Yep, this is correct, you can do this and would probably be the easiest
solution. However, the Classic SE data will not be accessible via the SRM
interface, nor will the remaining space in the NFS mounted filesystem. We
want to maximise the amount of space available. Also, it is a more
scalable solution to have gridftp servers running on each disk server,
rather than having to go through another node to get to the data. I think
this is the real reason why Simon didn't want to keep the NFS mount.
---
and Simon wrote:
---
Basically what I would like to do is
provide you with a perl regex (or similar) to translate the actual
location of the files, while I would move them as necessary myself behind
the scenes. So if you could come up with a script, as you say probably
similar to the classic SE migration, into which I can insert this regex,
and it will updated the database, that would do the job.
I really need this because we had create a new pool on the nfs-mounted
disk to get something working, but as Greig and Olivier explained the aim
is to do away with NFS and have a pool service on each of these
file server nodes.
---
The discussion was off list till now because it started as an
email exchange about something else :-)
So we need to understand whether the catalogues need to be
migrated. There are two potential alternatives:
1. The shutdown approach: tell experiments to copy what they
want to keep from the old SE to the new one. Problem is
both SEs will be using the same disk...
2. Renaming the disk server to make it the SE, as Greig suggested.
Cheers,
--jens
|