Andy, Recall that you POST the Estimate resource which refers to the Scenario resource. The Scenario describes one possible way to run a project (e.g. "as soon as possible" or "limit headcount to 8"). You then estimate metrics for a given Scenario and POST Estimate resources that contain probability distributions.
You could PUT an Estimate resource to modify it. The Estimate contains a probability distribution in some format, e.g. Triangular, Quantile, etc. One can compute the quantiles from a given distribution so you can assign a meaning to requesting percentiles for any Estimate, but that is really an operation the service would perform. A simple example would be to ask for the mean or standard deviation of any Estimate. That information may not be explicitly represented in the Estimate resource, but an application could define the meaning. The application would become a Statistics service. Perhaps a cleaner way to design this would be to have a new Statistics resource that accepted queries involving Estimate resources, and computed statistics for them. Arthur Ryman, IBM DE Chief Architect, Rational Project and Portfolio Management Office: 905-413-3077, Cell: 416-939-5063 Assistant: Nancy Barnes, 905-413-4182 Andrew J Berner <[email protected]> Sent by: [email protected] 01/18/2010 01:43 PM To Martin Nally <[email protected]> cc [email protected], [email protected] Subject Re: [OSLC] resource: storage vs representation--specifying variations in the form of the return from a GET Again, the answers may be resource specific. Martin, you're certainly right that you can't PUT a quartile representation and GET a percentile one. But in this particular case (but probably not in other examples, like the flyover), the resolution of that is you don't PUT the quartile or the percentile. You don't PUT the probability distribution either. It is derived from what in the OSLC estimation interface is being called a Scenario (yes, it's an overloaded term we know). You can GET and PUT the scenario, that creates the probability distribution, which is then available through a URL that other resources can link to and GET. Then my description is about what the result of that GET call should be and how the consumer specifies that, since the underlying mathematical representation is not tractable to "ordinary" consumers. Even in the quartile/percentile case, there's a glimmer of this issue. If instead of parametric techniques to derive the probability distribution, you directly PUT the percentile approximation of the answer, you could derive the quartile representation, but, as Martin points out, not the other way around (other than by making an assumption of segment wise linearity, for the mathematically minded---but that's not the point here). So you cannot have two independent URLs for them that both allow PUT and GET. Interesting side variant on my problem! Andy Berner Lead Architect, ISV Technical Enablement and Strategy IBM Rational Business Development 972 561-6599 [email protected] Ready for IBM Rational software partner program - http://www.ibm.com/isv/rational/readyfor.html |------------> | From: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Martin Nally/Raleigh/IBM | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | To: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Geoffrey M Clemm/Lexington/IBM@IBMUS | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Cc: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Andrew J Berner/Dallas/IBM@IBMUS, [email protected], [email protected] | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Date: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |01/18/2010 12:01 PM | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Subject: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Re: [OSLC] resource: storage vs representation--specifying variations in the form of the return from a GET | >--------------------------------------------------------------------------------------------------------------------------------------------------| I agree that accept headers are often the right solution, but it's less clear to me that it's the right way to handle the example that Andy brings up. Ideally, accept headers would only be used to choose between representations that are equivalent, like XML, JSON and N3. A quartile approximation of a distribution is not equivalent to a percentile approximation - if I use PUT to update the distribution through the quartile approximation, I will not subsequently be able to GET the percentile representation and retrieve a useful result. If I were doing Andy's example, I would create an URL for the distribution, and a separate URL for quartile and percentile approximations of it, rather than using content negotiation. the URLs might be <distributionURL> and <distributionURL>?samplingRate=4 and <distributionURL>?samplingRate=100 or something, By the reasoning above, you would conclude that we should not have used content negotiation for the "compact" resource format used for fly-over help (I forget the right name for these). If we were doing that over, I would recommend using a different URL, not accept headers Best regards, Martin Martin Nally, IBM Fellow CTO and VP, IBM Rational tel: (949)544-4691 |------------> | From: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Geoffrey M Clemm/Lexington/IBM@IBMUS | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | To: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Andrew J Berner/Dallas/IBM@IBMUS | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Cc: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |[email protected], [email protected] | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Date: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |01/17/2010 07:54 PM | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Subject: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |Re: [OSLC] resource: storage vs representation--specifying variations in the form of the return from a GET | >--------------------------------------------------------------------------------------------------------------------------------------------------| |------------> | Sent by: | |------------> >--------------------------------------------------------------------------------------------------------------------------------------------------| |[email protected] | >--------------------------------------------------------------------------------------------------------------------------------------------------| With HTTP, one requests a variant of the representation with the ACCEPT header. If one cannot encode the variant information in the ACCEPT header, I suppose one could define a URL parameter for it. Cheers, Geoff From: Andrew J Berner/Dallas/IBM@IBMUS To: [email protected] Date: 01/17/2010 06:04 PM Subjec [OSLC] resource: storage vs representation--specifying variations t: in the form of the return from a GET Sent [email protected] by: An issue that has come up in several of the workgroups I'm attending: The storage of a resource may be very different from the XML representation of the resource you GET from the URL. The spec defines the representation you GET and PUT, not the underlying storage mechanism. The XML representation specified in the OSLC spec may be derived by the server from the data stored. This raises a question about the spec: Have we been specifying ways of GETting alternate representations of the same resource? I'm wondering if there shouldn't be a somewhat standardized way to do this, although the specifics will be resource dependent. As I look over some of the specs, it seems we are defining a single XML format for each resource. In some cases, that may make sense--the resource IS the XML document. But in other cases, the resource is more abstract. Even if we expect the stream returned by an OSLC GET call to always be an XML document, there may be variations in the XML representation of a particular resource that could be important to be specified by the request. One example of this is will show up in the Estimation OSLC spec: the representation of estimated duration (or other key metrics) of a project. It's not a single number, but rather a probability distribution, so you can say the probability of completing the project in that amount of time. It's a formalized representation of something like this: "it's 50% likely we can complete it in 24 months, but if you want to be 75% confident, plan for 35 months and it's barely possible it could take over 4 years. There's a 25% chance you can get lucky and finish in 16 months, but the 12 month schedule you asked for is impossible! The least it will be is 14 months." Although "probability distribution" is an abstract mathematical construct, the XML representation that will be in the estimation OSLC spec is a quantile representation, which will be "good enough" for consumers. It will be an XML representation of some of the values of the probability distribution, but it can be at various granularities. It will list the values of the distribution for evenly spaced percentages as a sequence of values in the XML document--the client can figure out which percentage each represents based on the number of elements and the order of the elements. Here's an example of the informal description above, based on a similar example from the WIKI: <ems:QuantileFunction ems:metric="http://open-services.net/software-metrics/duration"; ems:unitOfMeasure="month" ems:numberOfQuantiles="4"> <ems:low>14</ems:low> <ems:quantile>16</ems:quantile> <ems:quantile>24</ems:quantile> <ems:quantile>35</ems:quantile> <ems:high>50</ems:high> </ems:QuantileFunction> Now, for some purposes, a client may want a representation of the distribution at a fairly high granularity, like the one above, at the 25%, 50% and 75% probabilities, maybe to get a ballpark of how risky the schedule will be. For other purposes, a client may want a representation at a detailed granularity, getting each percentage probability, perhaps to answer the question, "what's the probability I can finish by Dec 1?" and get an answer like "68%". I'll spare you the similar but much longer XML representation of the same resource (the probability distribution) but with 100 quantiles :-). This is just an example; I suspect each workgroup has examples like this, where there can be different levels of detail or other variations returned by a GET, all formatted by the interface provider from the same abstract resource. It's related, but not quite the same I think, as the standard issue of "partial retrieves" in queries (which I've started to hear described as the "shape of the returned set"). So there needs to be a way for the consumer to negotiate with the provider, or, more simply, just tell the provider how the consumer wants the resource represented, even if all the representations are in XML. Even though the "choices" of representation are resource dependent (e.g. "number of quantiles" for a probability distribution), is there a general technique we'll use in the various OSLC specs to specify requested variations? Andy Berner Lead Architect, ISV Technical Enablement and Strategy IBM Rational Business Development 972 561-6599 [email protected] Ready for IBM Rational software partner program - http://www.ibm.com/isv/rational/readyfor.html _______________________________________________ Community mailing list [email protected] http://open-services.net/mailman/listinfo/community_open-services.net _______________________________________________ Community mailing list [email protected] http://open-services.net/mailman/listinfo/community_open-services.net _______________________________________________ Community mailing list [email protected] http://open-services.net/mailman/listinfo/community_open-services.net
