Andrey, I think Denis meant that if there is no data for a provided ID on the node, query execution time will be close to nothing (of course, assuming that there is a proper index). Network utilization is minimal for these requests as well, because there are no results going back. In other words, most likely this will not slow you down, because you will wait for the slowest node anyway (in this case that's the node that has data).
But even this can be optimized with the approach I described earlier. You can use IgniteCompute.affinityCall() to send a closure to the node that stores the data for the ID and execute local query there. This way you avoid meaningless network trips to other nodes. Makes sense? -Val -- View this message in context: http://apache-ignite-users.70518.x6.nabble.com/Sql-query-performance-with-partitioned-caches-tp2700p2746.html Sent from the Apache Ignite Users mailing list archive at Nabble.com.