[ogsa-d-wg] Draft Charter for Data Movement Interface Standardization WG

[Malcolm Atkinson]

Hi Bill

I agree that such a standard interface is needed.
When you look at files I presume you consider files where ever they are,
secondary or tertiary storage at least.

When you say any dataa, then there is the possibility of trivial or
large amounts of data between RAM, as well as data from files and
databases with an enormous set of possibilities of the way it may be
selected and identified.  Eventually it gets close to Byte-IO, Streams
(BoF) and InfoD etc.

So I'm agreeing with you that if yu go beyond files then scope control
is difficult.

Would it be better to do the standardisation of file movement first and
look at other forms of adta movement later?

Malcolm
 

 >-----Original Message-----
 >From: owner-ogsa-d-wg at ggf.org 
 >[mailto:owner-ogsa-d-wg at ggf.org] On Behalf Of William E. Allcock
 >Sent: 14 September 2005 17:04
 >To: ogsa-d-wg at ggf.org; gsm-wg at ggf.org; byte-io-wg at ggf.org; 
 >Peter Kunszt; James Casey; Ravi Madduri
 >Subject: [ogsa-d-wg] Draft Charter for Data Movement 
 >Interface Standardization WG
 >
 >Sorry for the re-send, I typoed the byte-io mail list.
 >
 >All,
 >
 >Sorry for the SPAM, but I sent this to the "likely suspects" who might
 >be interested.  I have a proposed BOF (waiting for AD approval) to
 >discuss standardizing an interface for invoking data movement.  There
 >are several of them out there already.  CERN has the File Transfer
 >System (FTS), the gsm-wg has SRM copy, Globus has the Reliable File
 >Transfer (RFT) service, etc..  I don't think there will be 
 >any argument
 >that there is a need for such standardization, the hard part will be
 >scoping the extent of what we will work on.  For instance, all the
 >examples above are file based, but ideally, this interface would work
 >for any data that can be addressed.
 >
 >I expect that that the BOF will be centered around scoping the working
 >group, but I think we should (and approval of the BOF depends on)
 >getting some initial discussion around the scope.  So... here it goes:
 >
 >I think the obvious thing is that it needs to be able to have 
 >the basic
 >functionality presented by FTS, RFT, and SRM-copy, however 
 >the devil is
 >in the details, so I will break this up into "blocks of 
 >functionality":
 >
 >Lets start with naming.  What will this service accept as valid names
 >for entities that it will move?  URLs? EPRs? Will logical 
 >file names be
 >accepted or should they be translated outside this service?
 >
 >Related to the naming is what type of data will this service move?
 >Files? video streams?  the output of simulations? the output 
 >of database
 >queries?  Can we make this a service that any service that 
 >wants to move
 >data can simply invoke it?  Note that I am differentiating data from
 >messages.  You would not use this to send the result from a 
 >service that
 >summed a bunch of numbers, that would simply be a SOAP 
 >response... IMHO :-).
 >
 >Can we make a generic module that would allow this functionality to be
 >applied to any service that exposes the byte-io interface?  Does that
 >affect the interface or is it just an implementation issue?
 >
 >Can we make this service transport mechanism agnostic?  both 
 >application
 >transport (GridFTP vs HTTP vs ...) as well as network 
 >transport (TCP vs
 >UDP vs UDT vs ...).  My concern here is that I am not sure 
 >SOAP has the
 >functionality we need.  To do this, I wonder if we need the equivalent
 >of a union in C, so that the parameters specified are based on the
 >transport(s) chosen.  For instance, if you use TCP you need 
 >to specify a
 >buffer size, but not for UDP.  GridFTP specifies streams and data
 >channel authentication, but HTTP does not.
 >
 >What about security / authorization.  This is a broad category and we
 >should push as much as possible outside of scope via callouts 
 >and Policy
 >Enforcement Points (PEPs), but what about delivery guarantees 
 >such as AT
 >MOST ONCE, AT LEAST ONCE, EXACTLY ONCE, non-repudiation, etc.?  I know
 >Dieter has a set of use cases that require some of this type delivery
 >guarantee functionality.
 >
 >A potentially contentious issue is whether or not these services will
 >use WSRF and notifications to expose (push from the service) 
 >or methods
 >to query the state (pull from the service).  Hopefully, we can find a
 >way to make each optional.
 >
 >If we start making many optional parts to the interface, it will make
 >what is exposed as service metadata for brokering will become more
 >important.  I would propose that we should make at a minimum a
 >recommendation for what facts about the service should be exposed.
 >
 >All of the existing services accept "bulk" inputs, i.e., move 
 >these 100
 >files.  This can be a problem when the requests become very 
 >large due to
 >de-serialization.  Should we provide a "chunking" interface so that
 >requests can be of unlimited size?
 >
 >Please feel free to make comments on the above and more importantly
 >suggest other important issues we need to address.
 >
 >btw, once we have a mail list of our own we will quit 
 >spamming the other
 >lists :-).
 >
 >Bill
 >-- 
 >William E. Allcock
 >Argonne National Laboratory
 >Bldg 221, Office C-115A
 >9700 South Cass Ave
 >Argonne, IL 60439-4844
 >Office Phone:  +1-630-252-7573
 >Office Fax:      +1-630-252-1997
 >Cell Phone:      +1-630-854-2842
 >
 >
 >
 >