Craig Shallahamer:
1. Set based processing will likely be much faster than row based processing. Our experiment of processing 100K rows showed row based processing was 3700 times slower than set based processing. Not twice as slower or even 10 times slower… 3700 times slower!
（差别在3700倍）
2. Compared to set based processing, row based processing times degrade much quicker than set based processing. That is, row based processing does not scale nearly as well as set based processing. We saw this in that the linear trend line for row based processing was 0.00259 compared to 0.00000 for set based processing.
Row versus Set Processing, Surprise!（集合处理和单行处理数据的差异性）
原文：http://www.cnblogs.com/caroline/p/3961494.html