Eon Mode - Performance if data is not in the Depot and policy to rehydrate

Answers

• Options

  fxing Employee

  April 2020

  Hi,
  If the data queried is in the depot, especially for I/O bound queries, the performance is better than querying against communal storage location(when data is not in the depot). We have observed 2-3 times faster when comparing querying against depot vs. querying against communal storage for certain types of queries, this depends on data volume and query types, for other workloads, there might be more performance benefits if the data queried is in Depot.

  There are some scenarios that depot can be re-hydrated:

  • When a new node is added into the cluster or subcluster, depot warming may happen in background, Vertica will fetch data from communal storage by default.
  • When a query runs, but the data it references does not present in the depot, the query will run against communal storage first, under the over, there Vertica can fetch the data from communal storage to the local depot, once the data files are fetched into depot successfully, the subsequent runs of queries that reference to those data will benefit from depot.

  Hope this help, please feel free to let us know if you have further questions.

  Thanks, Fang

  0
• Options

  Jan_Soubusta Vertica Customer ✭

  April 2020

  2 - 3 x slower?
  Is this valid only for AWS (S3) deployment?
  I expect in case of Pure storage the degradation is negligible.
  Do you have any numbers for setup with Minio storage?

  0
• Options

  [Deleted User] Administrator

  April 2020

  In depot query performance is always going to better than communal storage in all Eon deployments cloud and in premise. Differences always depends on complexity of query and amount of data processed. Depot plays key role as performance booster for all Eon deployments and we recommend customers to take advantage of depot that is on local attached storage.

  0