jzhuge commented on PR #18336:
URL: https://github.com/apache/iceberg/pull/18336#issuecomment-6908784091

   @bryanck here are local JMH numbers (Spark 4.1, local FS, 10 files x 100k 
rows, ~690 MB; parent of 843d36de vs 843d36de). Bytes are rchar deltas from 
/proc/self/io per op.
   
   | Projection | MB read before -> after | s/op before -> after |
   |---|---|---|
   | id | 1.6 -> 1.6 | 0.118 -> 0.123 |
   | person.name (depth 1) | 36.2 -> 36.2 | 0.257 -> 0.274 |
   | person.home_address.city | 143.0 -> 36.2 (4x) | 0.294 -> 0.261 |
   | person.l2.l3.leaf (depth 3) | 263.6 -> 1.2 (222x) | 0.256 -> 0.179 |
   | person.l2, person.l2.x | 263.6 -> 263.6 | 1.128 -> 1.147 |
   | person (whole struct) | 584.1 -> 584.1 | 2.483 -> 2.491 |
   
   Before the fix, a deep leaf read exactly the bytes of its full intermediate 
struct. Whole-struct and struct+child selections are unchanged. On local disk 
the wall-time gain is small, because the extra columns were fetched but not 
decoded and come from page cache. On S3 the gain tracks bytes; that is where we 
saw ~8x on the 17 GB table. Mark's Cybertron numbers will cover that side.
   
   Benchmark code: #18985.
   


-- 
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.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to