Hi Pablo,

  well, I do not understand everything. This does not mean that I cannot 
help, I hope.

> We're building a Tier3 facility in the same place as our Tier2, that runs 
> dCache (with LCG software). The main idea is to use the olbd as a temporal, 
> volatile storage, caching the permanent data stored in the Tier2 MSS.

  So I asume that you are setting up a pure xrootd cluster in the tier3, 
interfaced to the MSS, am I correct?

> I'm new to this list, and wandering around the archives I couldn't find 
> accurate information about this. I've read many documents about the theory, 
> but not much about practice.
> I'm finding it really difficult to find good information to build such system 
> myself. So far the way to do this is via oss.msscmd and oss.stagecmd, that 
> require two different scripts/programs that perform the operation in the MSS.

  Yesm but the programs are furnished with the xrootd distro. In 
principle you can use your own, but in practice most people use the ones 
named "mps scripts", i.e. generic interfaces to a MSS-like system in the 
wider sense of the word. Wide sense because, for instance, in Alice we 
are switchig to use as a MSS another xrootd cluster. But the difference 
lies only in the config file (very tiny a difference) and in the mps 
extension scripts (mps.xfrcmd), i.e. the small scripts you have to write 
to interface your own MSS (e.g. HPSS).

> With the information in 
> one should be able to 
> build both, but there's not any example there or anywhere I've seen. Does 
> anybody here own a working example of both scripts and the xrootd config 
> file? It will save us a lot of time.

  I recognise that setting up mps scripts is not that easy, because the 
docs are exhaustive but quite complex.

  First of all I'd suggest that you set up the storage cluster without 
MSS, and you refine the config until you are happy. And then add to it 
the mps stuff. I can help with both phases, and if you go through 
official channels I can help you even more.

> Also, I've read there's a set of daemons that take care of deleting old data 
> to make space for the new, touching files that are being used, and so on. 

  The mps scripts (mps_MigrPurg) do such things, and by the way you 
don;t need to touch a file which is being used. This reminds to me a 
recent construction made by the PROOF guys, but it has quite a different 
scope. I'd keep that as a second step, making your cluster work is more 
important first imho, and then interfacing it to the mss comes second.

> Does it have any relation with xrootd, or anyone can lead me to more precise 
> and technical information about it? (also, access to the daemons themselves 
> would be nice)

  Don't worry about that. Setting up the cluster for normal work is easy 
typically. If you want, we can start from there.

> Thanks a lot,
> Pablo Fernandez