Notes

Items completed in this release:
1. Pegging can be off loaded. This is already done and it has produced 40% memory reduction. And this is for each scenario. We can use this memory to load other scenarios for Dell, PMI.
2. Development of Parallel DTO read framework is complete and delivered. For the start we are giving this feature for two reports
a. Short Alert Report
b. Push Alert Report
3. Pre-computed caching was in POC for this release. This is non-GA feature. We are using ehcache for caching results of the reports. It is proven and light weighted. It was previously only in web tier, but we are bringing to middle tier. It has capability to put in Heap, off-heap and disk or any combination of the three.
4. Caching pre computed comes with some risks, like, uncaching needs to be perfect and accurate, so the user is not seeing stale data after planning run.
5. We will be providing some super user level control to clear pre computed result cache, if things goes south.
6. We tested with 1 user, 5 users and 10 users retrieving at the same time using jmeter.
7. Due to this pre – computation, CBP run time has increased by 42 seconds, which is not bad and acceptable.
8. As POC, we implemented and took stats for couple of reports
a. Short Alert Report
b. Delivery Plan Report
Next Plan:
1. We will be benchmarking ehcache performance as well with 10 million records and test with different configurations of Heap, off-heap and disk or any combination of the three.
2. The total time to bring all the critical reports, which is around 15, extendible with pre-computed result caching feature is 1 release.
We will be implementing this feature only for planning reports not master reports.
3. If we do for all 15 reports and we are putting everything in heap, we will surely increase the heap memory.
4. Pre computation report and caching result is high priority item for 23.4. This will surely improve performance of cockpit as well.
5. Parallel DTO will be given to PS so that they can play around with it.
6. For JLR, currently running CBP with 2000 critical paths and 52 weeks planning takes around 40 mins.
7. We need to follow up line item to dig deep into CBP, to find root cause of long CBP run.
8. We can also use disk tier for storing cache results as we are using SSD for production boxes.
Open questions:
1. We can look for a way to parallelise the computation of 15 reports and putting in cache. This is will be asynchronous process. However, clearing the cache after planning run will be synchronous process.
2. Currently, we have ehache configuration at scenario level, can we have it at report level.? Code could get ugly and from a development point of view it is not good.
3. Can we have these ehache configured reports tested with more records, like for PMI or Dell. ?