[Linux-cluster] GFS block size

[Steven Whitehouse]

Hi,

On Thu, 2009-04-23 at 17:09 +0200, Mikołaj Radzewicz wrote:
> Hello again,
> I'm planning to run GFS cluster of two nodes on some SAN storage over
> fabric - 4TB ( 5 disk in RAID 5 - with stripe 128k). I'm having some
> mixed feelings about setting the GFS block size for my fs...
> 
> The nodes are going to hold some WWW files (let say for video streaming)
> which can be divided in 3 main groups:
> 1/3 files smaller 100K
> 1/3 files about 1M
> 1/3 files about 2M
> 
> I used to use a GFS block size of 4K - the default one but now I'm
> wondering of changing it to some greater value like 16K. Does it make
> sense and I will be able to notice some performance benefits?? On the
> other hand I have read that changing default 4K for GFS isn't advisable.
> 
> btw: if it makes some difference some directoris must contain about 50K
> files...
> 

4k is the largest block size available as we don't support block sizes
greater than the page size, so I'd stick with 4k for now. 50k files
should be ok in a single directory. Of more importance is the access
pattern (particularly creates and deletes) from the different nodes
within single directories,

Steve.