SAP HANA - does data storage block size impact IO so much?

other
1

I am doing SAP HANA performance testing.

I used the testing file size as 1GB, 5GB, 10GB and etc.
In each file size, I tested different block sizes for Log and Data - I tried 4k, 16k, 64k, 1MB, 16MB and 64MB block size.

The result is:

  1. the file size (1GB, 5GB or 10GB) doesn’t bring much difference.
  2. The block size shows very different results: when the block size is less than 1MB, the IO performances are very poor - the Write IO is even poorer than Read IO; when the block size is more than 1MB, the IO performance improves signicifantly.

I was wondering why the block size may impact the IO so much?
And, if using block size less than 1MB, what factors can I change to improve the IO performance so that it’s comparable with that of block sizes more than 1MB?

Thanks!

jameszhjw (BOB member since 2013-07-25)

2

[Moderator Note: Moved from General Discussion to Other Products]

And welcome to B:bob:B!

Marek Chladny :slovakia: (BOB member since 2003-11-27)

3

I don’t know about HANA but if it is analogous to a disk drive the block size has an impact on performance. A smaller block size requires more blocks to store a file. This in turn requires more reads and writes to the file allocation table so performance is degraded. With a disk drive you can optimize block size by calculating the average file size then make your block size as close to the average file size as possible. I have no clue if HANA works the same way though.

richardcottave (BOB member since 2006-03-30)