Write-block-size and max record size


#1

I am trying to understand the consequence of setting write-block-size, here is my understanding from official doc and answers in the forum, please correct me if anything is wrong. Thanks

  • Its default write-block-size is 1MB, for SSD it is recommended to use 128KB for better performance.
  • A record’s soft max size is write-block-size, it’s allowed that record size exceeds write-block-size with degrade of performance on writing/reading the record.
  • Even if a record’s content is only 10 bytes, when set it db will write the entire write block to SSD

Aerospike for large objects - LDT & LLIST (looking for alternatives to MongoDB and S3)
#2

This is what is stated, we are now finding that some SSDs perform better with larger write blocks such as 1MB. We recommend testing your SSDs with ACT. [Update] It turns out that this test result was flawed as the larger write-block-size had also increased the post-write-queue (8x) which is what caused the better performance.

No, the upper limit for the size of a record is the write-block-size writes exceeding this size will result in a error sent to the client.

No, a write-block can hold many records. Write-blocks contain record-blocks which are 128 bytes (non-configurable). A 10 byte record would use one 128 bytes record block within a write-block


#3

Thanks for the clarification!