[PVFS2-users] Re: I/O level
tblum at htwm.de
Fri Apr 22 01:54:18 EDT 2005
thanks again for your detailed answer. That was exactly why I was unsure
about that "object" thing. In reality the so called "object based" file
systems should actually called "ost based" fs.
So my classification for PVFS2 would be then "parallel file system using
objects for data organisation". Could you live with that? I won't let go
the objects ;-)
> Hi Thomas,
> That's actually sort of a trick question.
> In my mind, an object-based file system would store objects. Objects
> have explicit members, functions associated with them, associated
> semantics, etc. That's what you get in an object DB, for example.
> PVFS2 certainly is not that.
> In the storage arena right now, there's something called an "Object
> Storage Target" or "Object Storage Device", and there is an associated
> specification for interfacing to these OST/OSDs that is called T-10.
> The abstract reads:
> This SCSI command set is designed to provide efficient operation of
> input/output logical units that manage the allocation, placement, and
> accessing of variable-size data-storage containers, called objects.
> Objects are intended to contain operating system and application
> constructs." [ANSI T10/1355-D working draft]
> While there is a mechanism for associating attributes with OST "objects"
> in this model, it's certainly not objects as I think of them. Likewise,
> currently file systems going under the name "object-based file systems"
> are simply file systems built on top of OSTs. PVFS2 doesn't build on
> top of T-10 spec OSTs, so PVFS2 isn't one of these things either.
> However, PVFS2 is architected in a very similar manner to file systems
> built on top of OSTs: PVFS2 data servers manage local storage allocation
> on their own, present a model of variable-size data-storage containers
> (we call them data spaces), and allow for attributes to be associated
> with those data spaces.
> So you decide. I think we're mis-using the object-based file system term.
> Thomas Blum wrote:
>> Hello Rob,
>> thanks for your answer. The file level access is clear now but I'm still
>> bit confused because of the objects approach.
>> Can PVFS2 then be classified as an object based file system?
>>>PVFS2 provides user access, as you guessed, at the file level. This is
>>>actually identical to the approach used in file systems like Lustre,
>>>which thinks of the data stored on servers as "objects" but only
>>>provides a UNIX file interface to the data.
>>>Thomas Blum wrote:
>>>>I read the pvfs2-guide to find out at what I/O
>>>>PVFS2 does provide access to data. Am I right that it uses file level
>>>>access data? If somebody could explain or give a link to a description
>>>>that I would really appreciate.
More information about the PVFS2-users