toggle quoted messageShow quoted text
We compared the performance of both v1.4 and v.2.0 using public data, i.e., without collections. We have not observed any performance differences.
Since the introduction of pvtdata feature, the gossip coordinator checks whether any transaction in a given block modified a collection's data. If any transaction does modify a collection's data, the coordinator would check for eligibility, i.e., whether the peer org is a member of the collection. If the peer org is a member, the coordinator fetches the missing pvtdata from either local transient store or remote peers. If transactions in a block have not modified any collection's data (as in your case), the coordinator wouldn't try to fetch it.
In v2.0, this part of the code got refactored and some additional logs such as the one you have mentioned above were added. Otherwise, there is not much change in the logic. It is surprising to hear that you notice a drop in performance. We will look at the log statements and convert them to either debug statements or put behind an if-then-else as appropriate.
To debug the performance drop, can you share the value of the following metrics for both v1.4 and v2.0?
On Thu, May 14, 2020 at 3:26 PM <susheeldighade@...> wrote: