| Title | Hard to predict the mortality in a generation |
| Status | closed |
| Priority | optional |
| Assigned user | Gareth Rees |
| Organization | Ravenbrook |
| Description | When you set up a generation chain, you have to provide predicted mortality values for each generation [1]. This is very hard to do: on what basis can you estimate these numbers? And the numbers may not remain static over time. |
| Analysis | When the accounting reform work (job003772) has been merged, it should be possible for the MPS to measure the actual mortality of the portion of each generation that was condemned. (The MPS condemns zones, not generations, so it's possible that only part of a generation is condemned in a given collection. But it will still be possible to measure the mortality of that part.) The computed mortality could then be fed through a low-pass filter and the result used to drive the scheduling algorithm. This would alleviate the need for the client to specify the mortality of their generations. (Note that to support multiple traces we will probably need to maintain some information per pool generation per trace, say struct pertrace[TraceLIMIT] in the pool generation. Since TraceLIMIT is 1 at present this does not waste space.) |
| How found | inspection |
| Evidence | [1] <http://www.ravenbrook.com/project/mps/...topic/collection.html#generation-chains> |
| Created by | Gareth Rees |
| Created on | 2014-05-14 15:51:03 |
| Last modified by | Gareth Rees |
| Last modified on | 2016-09-05 16:53:29 |
| History | 2014-05-14 GDR Created. |
| Change | Effect | Date | User | Description |
|---|---|---|---|---|
| 192195 | closed | 2016-09-05 16:23:15 | Gareth Rees | Merge branch/2016-04-13/mortality. |