[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [wsrp] Re: Q about resource usage with http post
Thanks Rich. 1. I will need to look at Andre's feature proposal to be able to understand your comment. 2. The issue really is that for highly interactive applications, we need someway for the consumer to act as a proxy but then somehow pass the producer the control if needed as such. Consider the following business scenario as an example. a. A call center agent, let's call him Rich, logins to a CRM Customer Support Portal. The portal displays the user a page with the following 4 portlets (a hypothetical example) i. My Last 10 Open Cases ii. Open Cases from Platinum Customers iii. Escalations (if any) iv. Calendar The consuming portal essentially contacted the content provider CRM system(s) through WSRP and assembled the information by doing a getMarkup Operation on all of the portlets. b. Rich drills down on one of the cases from the list of 10 open cases. He does this by clicking on the link. The consuming portal acts as a proxy and calls performBlockingInteraction. The portlet in question maximizes its view and we see a page with the case detail for the case in question. c. Now Rich decides to update the case: (i) he updates the status --> an event fires that takes a server trip through the consumer. (ii) he updates the resolution --> another event fires.... another server trip through the consumer. (iii) he adds a few people to the notification list --> for each one of them, an event fires that looks up the individual and updates the list. All trip proxied through the portal. (iv) clicks on the notification button to notify the interested parties --> another event again through the consumer Note that i through iv are all performBlockingInteractions that if done according to the spec would result in the whole page being re-retrieved and re-assembled. Note that all a, b and c above need state information and thus by your definition have to be proxied. Consumer acts as a mandatory proxy for every request in the above use case. As a result, we see significant and in some cases prohibitively significant performance degradations. This is further aggravated if the consumer is striclty following the spec and actually re-retrieving (getMarkup) and re-assembling all the other portlets on the page. In my personal tests using just http gets and posts with a portal as the proxy (i.e. even without the overhead of SOAP over HTTP or the blockingInteraction calls), I have seen the degradation to be around 70-90% for the complete business scenario like the one outlined above. This was when all the portlets other than the one being interacted with were cached on the portal server. If you don't take caching into account as it is but optional, this figure would shoot up even more. How can I go to my CRM team and try to sell them WSRP. CRM products compete very aggressively on how much time it takes for an agent to open/close a case or change its status etc.. You can add more CPUs, more memory but its those extra hops due to consumer proxy that you have to jump over the network that kill performance. Not to mention the serial nature of performBlockingInteraction call. ----------------------------- I see this alongwith registry federation, federated identity mangement and authorization as the three major issues that we must resolve either as part of this specor other ones to make plug n' play reality. At PeopleSoft, we have been tacking these issues eversince we pioneered into the pure HTML era. We have a few ideas on how to resolve this. All of us in the WSRP committee need to make these issues our top priorities. I see this as being much more important than the less-core and more value-add features like interportlet communication. (On a lighter note) To start with, we have to stop taking the ubiquitous but entirely useless "Stock Portlet" as a use case. It is by far the most unrealistic and yet most frequently cited portlet by most portals that I come across. Other culprits are the weather portlet, Calendar, e-mail, clock and ofcourse the Dilbert portlet :-) Let's start analysing how WSRP can work within the context of critical business applications. Thanks, Khurram Mahmood Product Manager - Portal Solutions "Rich Thompson" <email@example.com To: firstname.lastname@example.org m> cc: (bcc: Khurram Mahmood/PeopleSoft) Subject: Re: [wsrp] Re: Q about resource usage with http post 09/27/2004 07:03 AM Thanks, this helped a lot ... my suggestions: 1. Resources are the matching concept as they are items directly requested by the browser for insertion into the overall page. Andre, I think this use case should be added to the related feature proposal you own and a number of requirements extracted from it. 2. The spec should clearly do a better job at calling out when a resource should be directly requested rather than Consumer-proxied. Times when a proxied request are needed include when the network configuration precludes direct access and times when the resource request needs stateful information that is likely to only be known by the Consumer. The generation of markup fragments for placement on the page certainly falls into the later case. 3. Resource requests need to be able to carry state information and request-specific information. The get/post issues need clarification and the semantics around any getPortletResource operation need to take these needs into account. Rich Khurram_Mahmood@peoplesoft .com To 09/24/2004 01:49 PM email@example.com cc Rich Thompson/Watson/IBM@IBMUS, firstname.lastname@example.org Subject Re: [wsrp] Re: Q about resource usage with http post I would like to emphasize the problem that Lars has described in detail below. In transaction-oriented highly interactive systems, such background interaction where only optimized pieces of the page are being refreshed instead of the whole page is increasingly becoming the rule instead of an exception. One may argue that the combination of HTTP+HTML was not designed to support such interaction but that's a moot point as we have to live within these limitation until rich clients bring back a smarter version of client-server. In the past few months, I have had discussions with quite a few content providers around the technical aspects of WSRP. More than a few have described it as a standard that is not fully baked out or is out of touch with reality. If you probe a little further one of the complaints that surfaces almost immediately has to do with the technical limitations that WSRP puts on highly frequent user interaction. Complaints ranging from mandating the consumer as the ubiquitous proxy essentially slowing down every single request to the implicit assumption that refreshing the whole page with every request would be acceptable to users in the real world. For example, in this case we will have to be forced to use the resource template and go directly to the content server whenever partial page refresh is involved. We don't want to do that and it is not an optimal design but the limitations of the spec will force us in this direction. Otherwise we don't support WSRP at all but that will not provide value to our customers and also significantly weaken the standard. It's time that the 2.0 version and subsequent specs focuss on the interactive systems and take requirements from them. We should try to accomodate those requirements in 2.0 and subsequent specs even if it means breaking the theoretically pleasing aspects of the spec. The bottom line is that the spec is useless until and unless content providers support it. We already have all the leading portal vendors supporting WSRP consumption but little non-trivial activity on production. What will make this standard viable and popular is the rapid increase in the number of content providers supporting it for production. Issues like the one that Lars has stated in his mail below present a significant road-block for achieving that vision. PeopleSoft is a leading provider of highly interactive enterprise applications. We will continue to work with the WSRP committee to make it into a more feasible standard for us and for other content providers who are facing problems that are similar to us. In my opinion, to enable wide adoption of this spec among producers, we will need a shift in the way we approach it -- a shift away from a portal vendor perspective to a interactive content provider's perspective. Without such a shift, we might end up with thousands of earnest consumers with nothing to consume. Thanks, Khurram lars_hofhansl@peo plesoft.com To: email@example.com cc: firstname.lastname@example.org, (bcc: Khurram Mahmood/PeopleSoft) 09/24/2004 10:03 Subject: Re: [wsrp] Re: Q about resource usage with http post AM We have potentially a lot of information to send; more than the maximum size of a URL (which is also browser dependent). We could chunk the information, send multiple GET requests and re-assemble the fragments on the server, but I'd rather perform a simple POST. There are also security issues. POSTed information is automatically secure when using SSL, for GET encoded data we'd have to do our own encryption (for example if access logging is enabled on the server where the resource is served from). -- Lars "Rich Thompson" <email@example.com To: firstname.lastname@example.org m> cc: (bcc: Lars Hofhansl/PeopleSoft) Subject: Re: [wsrp] Re: Q about resource usage with http post 09/23/2004 12:20 PM As I read your first paragraph, I also went to using the concept of resources as the right way to accomplish what you need (i.e. the updated fragment is a resource from the Consumer's point of view). Is there a particular reason the information you want to transfer has to be via http post rather than get? This sounds a lot like some of the things I did in a previous research project, but we used http get for all the transfers. Rich email@example.com 09/23/2004 02:35 PM To firstname.lastname@example.org cc Subject [wsrp] Re: Q about resource usage with http post In our case we are implementing a mechanims for "selective page refresh" using DHTML. I.e. we have to completely bypass the Portlet Interaction model. There seems to be no specific provision for this in the WSRP Spec (V1). For example performBlockingAction() either has to return the complete markup or it has to be followed by getMarkup(), we can't just return some change information and partially update a portlet. (I realize that when multiple portlets are displayed by a Consumer and one of the portlets needs to be re-rendered that the Consumer may re-render all portlet, which breaks our selective refresh paradigm for that case.) For these reasons we're trying to POST to resourceURL in order to handle our data exchange, then update the representation using DHTML based on the exchanged information without triggering any (Consumer visible) refreshes in the Consumer. Now, the V1 WSRP spec in 10.2.1.1.3.1 says that the cosumer is "encouraged to use the same communication style (e.g. HTTP Get or POST)" that was used by the user-agent. That does not seem to mandate that behavior and thus we cannot assume that all Consumers will indeed behave that way. Thanks. -- Lars Rich Thompson wrote: I don't know of cases where people have used http post in this manner, but the spec anticipates that such cases may exist and allows the markup to specify use of post with the requirement that the Consumer then also use post when passing the request on to the resource url. This keeps the Consumer truly acting as a proxy for these resources. Rich email@example.com 09/21/2004 05:36 PM To firstname.lastname@example.org cc Subject [wsrp] Q about resource usage with http post Is HTTP post supported for resource operations according to the spec? My reading of it points to an ambiguous statement to that effect on pg 62, section 10.2.1.1.3.1 Thanks To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to http://www.oasis-open.org/apps/org/workgroup/wsrp/members/leave_workgroup.php . To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to http://www.oasis-open.org/apps/org/workgroup/wsrp/members/leave_workgroup.php . To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to http://www.oasis-open.org/apps/org/workgroup/wsrp/members/leave_workgroup.php . To unsubscribe from this mailing list (and be removed from the roster of the OASIS TC), go to http://www.oasis-open.org/apps/org/workgroup/wsrp/members/leave_workgroup.php .