Re: Question for Provider Developers - It is possible to completely automate the process of writing a CMPI provider?

If the mapping document allowed for formulas, such as when a result property was 
actually derived from a calculation of several data source values, then perhaps 
it would be useful.
There would still need to be some stubs filled out by the programmer, but as 
long as all the basic stubs were generated, it would still save time and not be 
hard for the programmer to find and decide what needed final hand tailoring.

Being independent of underlying hardware is one of the problems.
While the client wants to be hardware independent as much as possible, the 
provider usually should not be.
In the case of hardware diagnostics, the goal was to have the vendor deliver a 
hardware specific diagnostic providers with each item they ship, and for the 
diagnostic to then produce a result generically interpretable by existing client 
diagnostic harnesses.  Even things that seem industry standards in hardware, 
such as SMART for hard drives, turns out not to be very generic.  While SMART 
creates a standard for hard drive self tests, each manufacturer uses different 
areas of the standard.  So a diagnostic provider for one brand of drive would 
have to evaluate different results criteria than a provider for another brand.  
But then all different brand providers could unify the result into a generic 
criteria, such as mean run time failure rate, or predicted lifespan.
I have not kept up with the latest profiles and standards, but I would assume 
there is still a lot not perfectly defined in the schema yet.

 Kirk & Trish Augustin
11821 NW McNamee Rd
Portland, OR 97231




________________________________
From: "Chatterjee, Biswapratap" <[email protected]>
To: "[email protected]" <[email protected]>; "[email protected]" 
<[email protected]>; "[email protected]" 
<[email protected]>; "[email protected]" 
<[email protected]>; "[email protected]" 
<[email protected]>; "[email protected]" <[email protected]>
Sent: Tue, November 23, 2010 2:39:42 AM
Subject: RE: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?

 
Hi all,
 
I thought of putting some light into what I am exactly planning for – 
 
	1. Please find below the      look and feel of the software. I intend the 
developer to only sketch the schema      diagram abstracting his underlying 
architecture. I have generated a sample      schema diagram using my “DCSGP” 
tool. J
	2. After completing the      sketch, the developer presses the “Generate Your 
Provider”      button which starts writing the CMPI provider C/C++ files, schema 
mofs and      registration mofs, by collecting information from the graph.
	3. The generated CMPI      provider will be using a Postgresql persistent 
database which will create      tables for the supported classes and 
associations (applying rdbms foreign      key concepts) by using information 
from the graph.
	4. The steps in point      number 3 will be done during the initialization of 
the CMPI provider.
	5. Later on when the cim      client fires queries like gi, ni, ei, an ,a, rn , 
r – those will be translated      into sql-queries and the result will be sent 
back to the cim client.
	6. The data source from      where the provider will collect information to 
fill up its CIMInstances      will be a “dynamic link library” with a generic 
interface like      – “map<data-type, data-value> getObjectInformation(      
<object-type>, <keys> )”.
	7. So in this way I      think the provider can become independent of the data 
source which it uses      to build its CIMInstances.
	8. A meta-information      like “a mapping document, xls sheet” when parsed 
will guide      the software to directly map data source properties with 
provider properties.
 
Problems:
 
	1. Not all values from      the data source are directly mappable to CIM 
Instance properties (contrary      to point 8). They need some conversions. 
Should the software ask the      developer to write explicit stubs in such 
cases?
	2. Executing methods in      SMIS? How generic they are if we strictly want to 
follow specifications?      i.e. if someone follows the specification for 
creating a volume, does he      become independent of the underlying hardware?
 
Regards,
Biswapratap.
 

________________________________
 
From:[email protected] [mailto:[email protected]] 
Sent: Tuesday, November 23, 2010 2:18 AM
To: [email protected]; [email protected]; 
[email protected]; Chatterjee, Biswapratap; [email protected]
Subject: RE: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
 
For those of you that have not seen the DMTF work in this area, I recommend that 
you look at: 
http://www.dmtf.org/sites/default/files/standards/documents/DSP2023_1.0.0b.zip 

 
What you will find is a much more comprehensive set of xsd’s used to define 
profiles.  The primary mission of  a profile is to define the service interfaces 
needed to satisfy a particular set of use cases.  The primary mission of a 
provider is to host the profile conformant interfaces and the code required to 
mediate between those interfaces and specific implementations.
 
Such implementations may sometimes be static, sometimes cached, and sometimes 
served dynamically.
 
As I see the problem outlined, it is to provide information about a particular 
type of implementation in a machine readable form so that a code generator can 
map between the profile specified interfaces and the mediation required to 
access the implementation.  More than likely, this work can/should be 
independent of the CIM server implementation, (i.e. Pegasus.)
 
George
 
 
From:Kirk Augustin [mailto:[email protected]] 
Sent: Monday, November 22, 2010 3:11 PM
To: Dodgson, David S; Steve Peters; Chatterjee, Biswapratap; 
[email protected]
Subject: Re: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
 
Sounds good but, "data items are populated by a separate service which monitors 
the state of our device and calls Pegasus when a change needs to be made", is 
not at all trivial.
It seems to me that this separate device is much more complex than providers 
themselves should be.
And it is sort of backwards.
The whole point of providers is to just act as a go between to allow client 
access to an unknown source of data.
The actual source of data should not be calling Pegasus, but the other way 
around, because Pegasus is not where the data is actually to be stored.
I think that to make providers easier to automate, one would want to make more 
assumptions about the various possible actual databases and persistent storage 
sources for properties, and to allow for the generation and configuration of 
default code to do the dynamic access and mapping.
 
Kirk & Trish Augustin
11821 NW McNamee Rd
Portland , OR   97231
 
 

________________________________
 
From:"Dodgson, David S" <[email protected]>
To: Steve Peters <[email protected]>; "Chatterjee, Biswapratap" 
<[email protected]>; Kirk Augustin <[email protected]>; 
"[email protected]" <[email protected]>
Sent: Mon, November 22, 2010 6:35:17 AM
Subject: RE: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
I presented a paper at the SNIA Storage Developer Conference entitled "Automatic 
Generation of SMI Providers".  It describes the method we use to automatically 
generate a provider based on the .mof files used by the Storage Management 
Initiative.  The paper is also supposed to be published with the Management 
Developers Conference that completed last week.  The SDC presentation should be 
available in one or two months.
 
In it I describe how we generate our provider from the .mof files.  We use 
SimpleWBEM along with a number of enhancements we wrote to do this.  The 
provider handles all data items but functions need to be coded individually.  In 
our case, the data items are populated by a separate service which monitors the 
state of our device and calls Pegasus when a change needs to be made.
 
This method may or may not be useful to others depending on how much 
customization is required.
 

________________________________
 
From:Steve Peters [mailto:[email protected]] 
Sent: Monday, November 22, 2010 2:22 AM
To: Chatterjee, Biswapratap; Kirk Augustin; [email protected]
Subject: RE: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
            Look at Cimple ( Simple Wbem). This comes very close to a tool to 
write a provider.
 
                        Steve Peters
                        PMC-Sierra
 

________________________________
 
From:Chatterjee, Biswapratap [mailto:[email protected]] 
Sent: Sunday, November 21, 2010 10:50 PM
To: Kirk Augustin; [email protected]
Subject: RE: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
 
Hi Kirk,
 
Thanks a lot for the reply.
 
I exactly thought in the same lines as you do, but then if the developer also 
had the option to share some information about its data source (binary from 
where the provider expects data); like a stub sitting in-between the data source 
and provider (providing a generic interface to gather data), and a mapping 
document (like an xls sheet) acting as a meta-information which directs the 
automation process to generate codes mapping provider properties to the data 
source’s properties.
 
I don’t know whether these things are possible. I am just throwing ideas, maybe 
we hit something. I just want to know the possibilities of such an idea.
 
Regards,
Biswapratap.
 
From:Kirk Augustin [mailto:[email protected]] 
Sent: Monday, November 22, 2010 12:04 PM
To: Chatterjee, Biswapratap; [email protected]
Subject: Re: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
 
And I think the answer is no.

Whenever I need to write a provider, I always start off with the same boiler 
plate basic provider, but there is no way to automate the creation of the rest.
Some is hardware specific, data persistent specific, protocol specific, relative 
to other machines, relative to processes and their states, etc.
HP had an SDK to make producing a provider much easily, but there will always be 
a need to write a significant portion of any provider yourself.
You can automate the created of a provider to satisfy a schema, but that only 
creates objects with storage space for property values.   It can't ever fill in 
the property values.  You always have to write the code to do that.
 
Kirk & Trish Augustin
11821 NW McNamee Rd
Portland ,  OR 97231
 
 

________________________________
 
From:"Chatterjee, Biswapratap" <[email protected]>
To: "[email protected]" <[email protected]>
Sent: Sun, November 21, 2010 9:36:05 PM
Subject: Question for Provider Developers - It is possible to completely 
automate the process of writing a CMPI provider?
Hi all,
 
This question has been banging in my head for quite some time. “It is possible 
to completely automate the process of writing a CMPI provider?”
 
I have been putting lot of effort in making this possible. Since, I have always 
believed that writing a provider is a repetitive task. If it is a standard,
then we have a set of guidelines to write it, we just have to develop an 
intelligent software who can understand the guidelines on its own and 

automate the process of provider development. The developer should only worry 
about the schema diagram that abstracts its 

hardware/architecture. Something like – “Draw your schema and get your provider 
shared object”.
 
If you feel this is an interesting topic than we can take this discussion to the 
next level about – “Possible ways to accomplish this objective”.
 
Waiting for your replies and ideas ….
 
Regards,
Biswapratap.