MemSQL 201: Tips(2)

that take advantage of encodings and hardware instruction sets to get performance gains.

Question: Why are Pipelines the best way to ingest data into MemSQL?

Answer: Within MemSQL Pipelines, each leaf node partition in MemSQL has its own “worker” process to ingest data in parallel, as opposed to inserting data normally through an aggregator node. This allows the ingest to leverage the distributed computing power of the cluster and to run concurrently, in contrast to traditional serialized ingest methods.

Question: How do the shard key and sort key differ in practice?

Answer: The shard key applies to every MemSQL table and refers to the distribution of data among the leaf nodes in a cluster. The sort key only applies to columnstore tables, and facilitates better performance by aligning the ordering of data on disk with the ordering in which it is likely to be accessed by queries. Hence, it is necessary to look at the most frequently run queries on your table when determining the ideal ordering of fields in the sort key.

Question: What is the difference between a broadcast and a reshuffle operation?

Answer: A broadcast operation moves a set of rows cleanly across the network to each of the other nodes to complete a distributed join. In a reshuffle (or repartition) operation, one side of the join is reshuffled (has its shard key recalculated with a different set of key columns) to match the shard key of the table on the other side of the join. Each respective operation is useful in different contexts when optimizing the join performance of a workload.

本文来自电脑杂谈，转载请注明本文网址：
https://www.pc-fly.com/a/jisuanjixue/article-81538-2.html