Hi Volodymyr,
For the posix access, how difficult would it be for dCache to expose
its cache to xrootd (on the same servers) via a server-local soft link
namespace to files on disk, for example? (Such a thing _does_ work as the
Wisconsin people did this at one point to serve data in dCache via xrootd,
although they had to create this server-local soft link namespace by hand.)
Pete
On Mon, Sep 19, 2005 at 04:26:50PM -0700, Andy Hanushevsky wrote:
> Hi Volodymyr ,
>
> Hmmm, seems that when I go down this path, there appears to be a disconnect
> between the SRM and a data access system concepts. What xrootd provides to
> clients is low latency high performance random access on par with the best
> commercial systems out there. That can't be done with the SRM protocol
> (i.e., delivering random access data to client analysis jobs). On the other
> hand, xrootd relies on some other entity to manage the disk cache. All that
> xrootd requires is physical access to that cache and some amount of
> predictability in its management (though it's reselient to crazed
> management). To date, xrootd interfaces to the mps system which can be
> customized to talk to a general purpose MSS (it can actually talk to an
> SRM/DRM combination as well if you don't mind double caching).
>
> The general integration of SRM calls for using SRM facilities to bring
> files into some disk cache when needed. Access to data in those files uses
> either the root framework or the Posix calls (which in turn use the
> xrootd). Where that fits in depends on what whether the SRM implementation
> provides some disk cache management features or not. So, from an SE
> perspective, the SE provides an SRM interface to the outside world but
> internally provides xrootd access to applications.
>
> Andy
>
> ----- Original Message -----
> From: "Volodymyr Kapchynskyy" <[log in to unmask]>
> To: "Andy Hanushevsky" <[log in to unmask]>; <[log in to unmask]>;
> "Patrick Fuhrmann" <[log in to unmask]>
> Sent: Monday, September 19, 2005 1:28 AM
> Subject: Re: dCache wants to communicate with XRootd
>
>
> >
> >Hi Andy,
> >
> >Thanks for the detailed reply.
> >We will make ourselves familiar with the descriped
> >mechanisms.
> >There is just some question left about integrating
> >SRM into the xrootd service.
> >Did I understand correctly that
> >"to integrate xrootd with a generalized SRM" means that
> >instead of having oss/ofs and/or MPS you would prefer to talk
> >SRM client protocol and with that have a general interface
> >to mass storage systems (disk/tape) ?
> >BTW: are there any plans to let xrootd talk the SRM
> >server protocol so that it can serve as a Storage Element ?
> >
> >dCache Team
> >
-------------------------------------------------------------------------
Peter Elmer E-mail: [log in to unmask] Phone: +41 (22) 767-4644
Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland
-------------------------------------------------------------------------
|