ext3 block size

ext3 block size

Post by Dani Oderbo » Sun, 10 Aug 2003 00:39:46




Quote:> hello.
> my database size is 5GB. what is the block size recommend?
> thanks
> wilson

Hmm, IMHO this depends more on your IO System
rather than on the DB size.        
It doesnt make sense to have a blocksize which your IO System
(HD, Controller...) cannot fetch in one read.
And it also depends on the type of Processing you are doing.
If you will have mostly big tablescans, a big blocksize is good,
whereas if you will be doing lots of Index acess, a lower one
can make more sense.
(And also if you store lots of data in one row).
In general I guess 8K (the default) is ok,
but 16K can make sense.

Regards,
Dani

---------------------------(end of broadcast)---------------------------
TIP 2: you can get off all lists at once with the unregister command

 
 
 

ext3 block size

Post by scott.marlow » Sun, 10 Aug 2003 00:51:56




> > hello.
> > my database size is 5GB. what is the block size recommend?
> > thanks
> > wilson

> Hmm, IMHO this depends more on your IO System
> rather than on the DB size.        
> It doesnt make sense to have a blocksize which your IO System
> (HD, Controller...) cannot fetch in one read.
> And it also depends on the type of Processing you are doing.
> If you will have mostly big tablescans, a big blocksize is good,
> whereas if you will be doing lots of Index acess, a lower one
> can make more sense.
> (And also if you store lots of data in one row).
> In general I guess 8K (the default) is ok,
> but 16K can make sense.

But that's confusing postgresql block size with ext3 block size.  Right
now the linux 2.4.x kernel only really understands 4k block size as the
max, due to limitations of architechture (X86 only understands 4k memory
page size, and the max block size of a file system is the max page size of
your architecture.  Now, while Itanium (8k) and Itanium2 (64k) can have
larger page sizes, not much has been done to make the 2.4.x series
kernels use these larger limits on hard drive blocks, and the little bit
of hacking I've found on it makes it look like it's still something that's
being tested, i.e. it's not ready for prime time.

Switching to Postgresql block sizes...

Note that while Postgresql supports larger block sizes, it isn't
tested as much as the 8k block size, so you might want to test it
thorougly before putting it into production.

---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

               http://archives.postgresql.org

 
 
 

1. OS block size vs. DB block size

The blocksize on your disk is 512 bytes, so if you're running RAW you want
to make it multiples of that.  If you're running through standard UNIX
filesystem then you want to make it a multiple of what the UNIX kernel has
as its blocksize.  I thought Solaris was a 4K block for its kernel buffers,
who told you it was an 8k kernel block?
Thor HW

2. Oracle -> MS-SQL Server ?

3. Oracle block size - OS block size

4. merging 2 databases

5. pg_dump 2 gig file size limit on ext3

6. Indirect vnodes

7. DB Block Size Revisited + VxVm Stripe Size(newbie)

8. NT Cluster size and Oracle data block size

9. RAID stride size (stripe) and Oracle block size

10. blocks, blocks, and more blocks

11. blocking, blocking, blocking agghh!

12. RAID block size vs. NTFS block size?