Alexander, January 18, 2013 - 1:31 pm UTC
Why does it not matter that the Oracle block size match up with the default Unix (Linux at least) block size? I would think it would be beneficial to have them line up from an OS memory management point of view. I don't actually know what I'm talking about but it just sounds right and I wonder what the explanation is whether it is true or not.
January 18, 2013 - 1:57 pm UTC
linux block sizes are typically 1k (sometimes 2k, 4k), so they all line with up our block sizes nicely.
Alexander, January 18, 2013 - 2:25 pm UTC
I thought Linux (64 bit) was typically 4k. Oracle being 8k, wasn't sure what the relationship was there.
This reminds me of something else I wanted to ask on a related subject, have you ever in your life went to using hugepages to solve or improve any Oracle performance problems?
January 18, 2013 - 3:59 pm UTC
Thankyou
Doug, January 20, 2013 - 6:10 pm UTC
Thanks Tom
with securefile lobs?
Charles, January 29, 2013 - 6:25 am UTC
Hi Tom,
Would you take the same stance of universal 8k block size even when using securefile lobs in their own tablespace? Have you seen any benefit from a 32k block size for lob tablespace and 8k block size for tablespace holding the structured data?
thanks for your opinion on this.
Charles
January 31, 2013 - 9:19 am UTC
there could be some minor efficiencies to be seen by aligning the block size with the chunk size - but not really. LOBS are managed in chunks already, not in single blocks necessarily.
I have not directly measured this for a few releases, it would be something to benchmark and evaluate. but I'm leaning towards "no"
ik
kylsij, February 17, 2013 - 4:34 am UTC
ik
A reader, July 16, 2013 - 11:17 am UTC
db block size for large ETL/DW environment
Ravi B, February 03, 2014 - 9:50 pm UTC
16K blocksize
Roger MacNicol, November 09, 2017 - 1:58 pm UTC
The comment about Oracle not testing 16k blocks is just wrong. A quick scan of the regression tests shows a very large number running on 16k blocks and Oracle's DW stress tests run on 16k blocks.
That said, worrying about block size is like fiddling while Rome burns - many (most?) apps have low hanging fruit which will give a better ROI on your time than tuning the block size.
8k is still best block size to be using.
November 09, 2017 - 2:29 pm UTC
Thanks for letting us know Roger.
32K block size regression testing
Naresh, March 20, 2018 - 2:31 pm UTC
hi,
Thanks for the info on 16K regression tests.
Do you also have the same level of regression tests done for 32K block size?
March 21, 2018 - 11:14 am UTC
That said, worrying about block size is like fiddling while Rome burns - many (most?) apps have low hanging fruit which will give a better ROI on your time than tuning the block size.
8k is still best block size to be using.