Github user xuchuanyin commented on the issue: https://github.com/apache/carbondata/pull/2611 @kumarvishal09 can you explain this modification? In previous implementation, we split a record to 'dict-sort', 'nodict-sort' and 'noSortDims & measures'. 'noSortDims & measures' is packed to bytes to avoid serialization-deserialization for them during reading/writing records to sort temp. In previous implementation, we can see about 8% enhancement in data loading.
---