PI Data Archive: Improve piarchss performance
PI Archive Subsystem should be able to use operating memory more efficiently. For example by loading as many non-primary archives into its process memory as possible (perhaps it could be affected by a user setting).
Could you please provide more details and examples of what you want to do? The Archive Subsystem already manages the files in memory, and in the file system cache. So, what are your specific situations where you see inefficient memory usage today?
Jinmo Yi commented
The windows file system cache currently take files off of memory if it's less used. When system has very large memory(>100GB), in some cases, Windows will voluntarilly free memory as some archives are less used. This is not efficient for the occasional analytic purpose long term queries. It would be nice to be able to set "keep at least x months of data in cache"
Joey Langan commented
I'd like to add something here. I'd like to see where I could force certain archives into memory. While I agree that the archive ss does manage the memory along with the file system cache, there is no way to override it given the use of the system. Ideally, a certain portion of memory would be reserved to say keep the last x days archives in memory. But right now, that is not forced. So if I have users doing very large searches, they end up dumping some of the last x days out of memory. I should be able to pick archives that are memory resident depending upon what my users are doing (e.g. I always want to make sure that the last x days are loaded into memory so that my analysis server and users have the quickest response, or maybe a user is doing a study over a time period and wants those in memory during pull of the analysis). Currently there is no way to force the caching based upon business requirements.