Hi, Thank you very much for your response.
The solution for the 1st problem is reasonable, but I have some concerns about the second one. In an autonomic computing environment, system attributes are monitored periodically and some actions are executed as the result. In other words, the policies are evaluated at different intervals (e.g. every second). So, it would be so expensive and repetitive to parse/validate the policy for every evaluation, and it degrades the system performance. Another approach could be just to check the policy syntactically at parsing phase (not semantically) and assume that the referenced attributes and methods exist at evaluation time (i.e. semantically is fine), and if the mentioned attributes are not found at that time, an exception can be thrown. So the only difference with the existing approach is that the developer gets the exception at evaluation time rather than parsing time, and it is not a big deal. However, we would like to participate in resolving these issues and please let us know how we can contribute to the project. Thanks again, Best regards, Reza _____ From: Prashant B Baliga [mailto:[EMAIL PROTECTED] Sent: May 12, 2008 8:27 AM To: Reza Asadollahi; [email protected]; [EMAIL PROTECTED] Cc: David L Kaminsky Subject: Fw: Autonomic Computing with Imperius Hi, I propose adding a new method to the Java-SPL as below to extend the policy infrastructure for dynamically generated classes/objects/beans. We could add a method that takes in the java object/bean and the policy string. This method would use string manipulation to add the import statement which it would get from the object. The parsing/validation and evaluation could become a single flow here to accommodate dynamic evaluation. Let me know if you have any suggestions. Thanks & Regards, Prashant Baliga Autonomic Computing Policy Development, India Software Labs, IBM Software Group Tivoli, IBM. http://incubator.apache.org/imperius/ ----- Forwarded by Prashant B Baliga/India/IBM on 05/12/2008 17:44 ----- Neeraj Joshi/Durham/[EMAIL PROTECTED] 05/08/2008 18:30 To Prashant B Baliga/India/[EMAIL PROTECTED] cc Subject Re: Autonomic Computing with ImperiusLink <Notes://d23ml170/652571910080569B/38D46BF5E8F08834852564B500129B2C/3959D6A5 61B9427A652574390041D355> My comments in red Sorry was on vacation hence the delay in replying Neeraj ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ "The light at the end of the tunnel...may be you" Neeraj Joshi WebSphere XD - Compute Grid AIM, IBM Apache Imperius - http://incubator.apache.org/imperius ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Prashant B Baliga/India/[EMAIL PROTECTED] 04/29/2008 01:32 AM To Neeraj Joshi/Durham/[EMAIL PROTECTED] cc Subject Re: Autonomic Computing with ImperiusLink <Notes://D03NM118/87256CAF007435F0/38D46BF5E8F08834852564B500129B2C/FABA6323 36E67F3165257433004A8EB2> Hi Neeraj, My comments are in blue below. Thanks & Regards, Prashant Baliga Autonomic Computing Policy Development, India Software Labs, IBM Software Group Tivoli, IBM. http://incubator.apache.org/imperius/ Neeraj Joshi <[EMAIL PROTECTED]> 04/22/2008 18:54 Please respond to [email protected] To "Reza Asadollahi" <[EMAIL PROTECTED]>, [email protected], [EMAIL PROTECTED] cc David L Kaminsky <[EMAIL PROTECTED]> Subject Re: Autonomic Computing with Imperius Hi Reza, Thank you for your interest in trying out SPL your comments are greatly appreciated. > For example bean.attr1 should be translated internally to bean.getAttr1(). I think this is a good idea and do not think it will be very hard to implement. Getters and Setters are not mandatory in Java. We would need some logic to see if the method exists etc. Is this how we should implement it? We currently allow access to private variables through getters and setters right? and it can be used outside the condition block. attr1=bean.getAttr1(); What we could do is first check if bean.attr1 is accessible (ie is public) if not we can try doing getAttr1() that way we remain flexible >anchor classes are kind of dynamically generated classes, and I can not specify their class names This one is a bit tricky because a policy has a 2 phase life-cycle in phase 1 the policy is validated syntatically and semantically and an *executable* is generated and cached. In phase 2 the policy is evaluated using the input parameters. Not knowing the class name in phase 1 will naturally be a problem so we would have to be able to support a way to evaluate on the fly...this may require some fundamental changes. We could add a method that takes in the java object/bean and the policy string. This method would use string manipulation to add the import statement which it would get from the object. The parsing/validation and evaluation could become a single flow here to accommodate dynamic evaluation. Yeah thats a good idea We are always looking for folks to participate in our project so if you have the cycles you could just provide us with a patch for #1 and I can check it in. I have cc'ed this email to the community for others to comment Thanks! Neeraj ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ "The light at the end of the tunnel...may be you" Neeraj Joshi WebSphere XD - Compute Grid AIM, IBM Apache Imperius - http://incubator.apache.org/imperius ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ "Reza Asadollahi" <[EMAIL PROTECTED]> 04/21/2008 04:07 PM To David L Kaminsky/Raleigh/[EMAIL PROTECTED], Neeraj Joshi/Durham/[EMAIL PROTECTED] cc Subject Autonomic Computing with Imperius Hi, I am doing research in the area of autonomic computing in the University of Waterloo Canada. As part of my research, I was intending to develop an implementation of CIM-SPL, and hopefully, I found that you did it. Thank you so much. You did a great job, and I guess Imperius can be the leader of policy engines in this area. I used the current version of Imperius in my project and I found a few limitations in the application. I would like to share them with you and get your suggestions about them. 1. A policy can not access the anchor class private attributes via their getter and setter methods, and the attributes should be defined "public" instead. Yes, I found that it is possible to call a getter method in the "condition" section but it is more interesting to read and write attributes in an expression-like language. For example bean.attr1 should be translated internally to bean.getAttr1(). The idea of Java EE Expression Language in terms of using high-level expressions instead of java-like code could be helpful in this context as well. 2. The objects that I pass to the policies as anchor classes are kind of dynamically generated classes, and I can not specify their class names in the policy file (they don't extend a specific super class or interface). Hence, it is not possible to work with the anchor objects in the policies as sensors and effectors. For example, I have a dynamically-generated java bean and I know that it would have have an attribute named "attr1", but I don't know its class (and it is not important at all); however, I would like to able to use that java bean in the policy. To overcome this problem, I guess it would be enough for Imperius to work with the passed java beans by reflection. I believe that addressing the mentioned concerns in Imperius will significantly improve its feature set and facilitate its usage in different adaptation environments. I would like to thank you for your attention to my points, and I am looking forward to hearing from you soon. Best Regards, Reza Asadollahi
