FYI, and some of us (Andy, me and possibly even Wilko) will likely have to spend time dealing with it.
--
Wei Yang | [log in to unmask] | 650-926-3338(O)
________________________________
From: Rob Gardner [mailto:[log in to unmask]]
Sent: Friday, August 15, 2008 11:15 AM
To: Alain Roy; Yang, Wei; Tanya Levshina; Ted Hesselroth
Cc: Ruth Pordes; Michael Ernst; Rob Gardner; Andrew Hanushevsky; Alex Sim
Subject: Re: Bestman-xrootd documentation, installation
Notes from today's meeting on Bestman, Bestman-Xrootd, and
Xrootd integration in VDT. Comments/corrections welcomed.
regards,
Rob
--------------------------------------------------------------------
Attending: Wei, Rob, Tanya, Ted, Alain
Overview discussion
==============
We discussed the requests from ATLAS for including
"bestman-xrootd in VDT", and expectations for OSG-Storage.
There was general agreement this consists in three
parts, and we finally I think cleared up problems
with the nomenclature:
1) Bestman, which provides an SRM endpoint to
Posix file systems.
2) Xrootd - the storage system itself
3) Three pieces needed to connect Bestman to
Xrootd which are: XrootdFS, FUSE, and XrdCNS.
Bestman-Xrootd was (is) an separate implementation
of Bestman developed by LBL(Alex) and SLAC(Wei)
specifically to meet ATLAS requirements though the
resulting product is completely generic. It is a subset
of the full Bestman, and does not support space token
reservations in the most general sense. It does
implement a concept of static space tokens. It is also
10 times faster handling SRM requests. More info here:
http://wt2.slac.stanford.edu/xrootdfs/bestman-xrootd.html
Bestman-Xrootd has been integrated into the
original Bestman, and now is available as
a configuration option of Bestman called
"Bestman-Gateway". This is not yet in VDT.
Xrootd is already in VDT, though its an older
version.
The key pieces for integrating a new Bestman
to Xrootd are XrootdFS, FUSE and XrdCNS.
This is where most of the questions, from the
point of view of OSG-Storage, arise.
XrootFS is a package that Wei wrote which
implements a number of Unix file system
commands for the files in xrootd in a
Posix-like way that Bestman deal with, see
http://wt2.slac.stanford.edu/xrootdfs/xrootdfs.html.
This is done using FUSE, http://fuse.sourceforge.net/,
which comes with Scientific Linux 4, 4.5, 5
but otherwise can be installed by admins
from the website, and also XrdCNS.
It was agreed to best let admins handle the
FUSE piece themselves but provide pointers
as appropriate.
XrdCNS - the service which provides a composite
namespace (a directory of shadow files for
all of the files in an xrootd system) is needed
by XrootdFS and is currently provided
separately from xrootd. Wei says this
will eventually be part of the standard
xrootd release, in a few weeks time. For
now it was agreed to provide a package
for it in VDT.
Packaging + Configuration
==================
We discussed what makes sense here - it was
agreed to start simply, using the VDT as an
integration point for all the components
except FUSE: latest Bestman, xrootd,
and the integration piece (XrootdFS
and for now XrdCNS). OSG-Storage will
have further discussion and consideration
of the main packaging approach, rpm vs
pacman.
How a complete system gets deployed is a
major consideration from the start. Like
dCache, a number of servers will typically
be required and what bits get installed
on each depends on its role. Configuration
does not necessarily need to be fully
scripted, and indeed this is not practical
given the complexity. It was agreed
that based on test-stand experiences by
OSG-Storage (and later by VTB) general
guidelines would be provided for assigning
roles and installing & configuring packages.
It was agreed to start out simple and to
organize packages to allow maximum
flexibility for various types of sites. Again,
the storage team felt more experience
was needed here, and will be in contact
with Wei regarding recommended
config options for the test-stand.
Support
======
We discussed a bit the expectations for
support, what it really means. From a high
level I said ATLAS expects these components
to be treated as 'first class citizens' in the OSG
storage group along the same lines as
dCache. Expanding for the minutes here,
I think this means, for example, providing
the integrated set of components
that site admins can install from a single
reference point: packages in VDT as appropriate,
institutionalizing install procedures, and
to provide the OSG liaison to the Bestman
and xrootd projects as appropriate when
dealing with integration problems or software
bugs. (It does not mean fully automating
an installation which is likely too difficult
in the general case I think).
Regarding testing, I will ask the OSG Integration
group to use the VTB to test candidate releases,
as we have done with the dCache SE and other
services in OSG.
On Aug 14, 2008, at 1:17 PM, Rob Gardner wrote:
Alain,
thanks. Can everyone confirm the following for a
meeting tomorrow:
11 CDT
Phone: (605) 475-6000, Access code: 735188
Agenda will be to talk through Wei's points focusing
specifically on VDT integration and installation
issues/dependencies.
thanks,
Rob
On Aug 14, 2008, at 1:12 PM, Alain Roy wrote:
I think it's a good idea, and that time is fine with me. Just let me know the coordinates.
-alain
On Aug 14, 2008, at 10:58 AM, Rob Gardner wrote:
cc'ing Alain here - I think smaller meeting might be
very useful. Does everyone agree? If so how
about tomorrow at 11 am CDT? (I can setup
the phone)
regards,
Rob
On Aug 14, 2008, at 12:42 PM, Wei Yang wrote:
Hi Rob,
I was not in that meeting but Alex Sim forwarded something to me, and
I e-mailed back to Tanya saying that xrootdfs is still needed despite the issue of FUSE kernel module.
Let me know if you want me to schedule a meeting with Alain, Tanya? I don't have Alain's e-mail address though I receive e-mails from him almost every week:-)
regards,
Wei Yang | [log in to unmask] | 650-926-3338(O)
Rob Gardner wrote:
Hi Wei
Let me chat first with Alain. I noticed b-m, xrootd was
discussed heavily in yesterday's osg-storage meeting.
Perhaps you're already in contact with Tanya?
I don't want to create more noise, but it seemed
to me a meeting between you guys and OSG (specifically
Alain and Tanya) would be helpful to get answers
to your questions.
Rob
|