Hi, looking for an advice. I'm facing an implementation of several contributor apps with elists longer then 2k elements.
Would you have any experience / comments with such apps? What would be the pitfalls here?
/chri5o
Hi.
Slow reconciles (unless you have a very big cluster) is the only problem, in my experience. Just don't add any total elements (they are recalculated every reconcile and with 1k elements it unbeleivably slow).
Why are you choosing such design?
If you have parents in the elist hierarchy, try not to exceed 20 children or so per parent (reconciles open up all data blocks for aggregration). Conversely, too few children per parents can increase save/submit time as you would get more lockouts during real-time aggregation up the hierarchy.
Buy a couple of dedicated job servers. Try to avoid cut-down models as that increases GTP time.
;D
Thanks for your comments.
Ykud, the design is unavoidable in this case, unfortunately we must follow this path. Several dozens of financial metrics on elist with 30 customers per each.
Elist is actually DB generated.
The app is today less then 300k per elist slice and reconcile of all app takes 6 min ;-) 1700 elist items.
We today have 14 threads in the cluster for reconcile jobs. Production env should be twice that.
GTP itself is not of much importance in this case, we will do it once a month or so. Even thou, cutdowns are switched off.
The thing I'm worried the most are the assess tables, but that also seems to work fine,...for now,...
6 mins is really nice. No data in applications? I'd recommend doing a stress-test: loading all elists with random generated data (that's easy to do in Analyst) and then measuring gtp time and response.
Considering ATs: you should avoid ATs with elist, if that's possible. ATs generating more that 6mb txt file can cause errors while import (had such experience).
For what is this application being used?
It almost sounds like it's reporting financial metrics, in which case Cognos EP is probably not the preferred solution. BI may be a better tool.
Is there any user data input required? Is the tool manipulating the data in any way aside from aggregation?
If the answer is, "yes, and it needs to stay in EP", then could you break into two or more models?
Hi, thanks for replies, the app is for data entry only.
Further on BI and some ETL is used allocations and reporting.