gianm opened a new pull request #109: Cache arrayBaseOffset. URL: https://github.com/apache/incubator-datasketches-memory/pull/109 I've been doing some investigations into DataSketches HLL performance in Druid (see also: https://github.com/apache/incubator-druid/pull/8194). One thing I found is that when the Druid query broker — a process responsible for merging partial results originating from each Druid data server — is merging results, it spends nearly 50% (!) of its CPU time calling `Unsafe.arrayBaseOffset` as part of `Memory.wrap` of serialized `byte[]` sketches. This patch caches those offsets, which speeds up the Broker code path by nearly 2x. <img width="1665" alt="image" src="https://user-images.githubusercontent.com/1214075/62093271-8835a400-b22d-11e9-8e97-764585f65030.png";>
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
![https://user-images.githubusercontent.com/1214075/62093271-8835a400-b22d-11e9-8e97-764585f65030.png"](https://user-images.githubusercontent.com/1214075/62093271-8835a400-b22d-11e9-8e97-764585f65030.png"){kind=link}