Servers - General
cancel
Showing results for 
Search instead for 
Did you mean: 

What block size should be set ??

 
xam
Occasional Advisor

What block size should be set ??

My file server will serve about 40 clients. And each clients will get large file (30Mb) in all the time.

What stripe size should be set for the optimum performance? 64K? 128K? 192K?

(I am using 6400 array card)

Thanks
8 REPLIES 8
Martin Balaj
Frequent Advisor

Re: What block size should be set ??

I think 64K is optimal for you .
xam
Occasional Advisor

Re: What block size should be set ??

Why 64? not 128? 192? 256?

Any formula to calculate? or by gusses?


thanks
Martin Balaj
Frequent Advisor

Re: What block size should be set ??

There isnt any calculation ,just meaning of using ,larger block sizes are good only for very big arrays with large databases an so on .Sorry for my english .
xam
Occasional Advisor

Re: What block size should be set ??

That mean even 128 or 256 are suitable?

I am trying a way to find the best setting for my servers.


: )
Martin Balaj
Frequent Advisor

Re: What block size should be set ??

You may use it ,but 30MB is not large file (or files)and therefore ,i think you have better performance with 64K .
xam
Occasional Advisor

Re: What block size should be set ??

ar....

anyway, thanks for your information
Thomas Bianco
Honored Contributor

Re: What block size should be set ??

64 is almost always the perfect raid block size. why? because thats the size the os expects.

the only time you would create a larger block size would be if your going to use a raw database partition, like oracle or... well i can't think of any others, but i'm sure most enterprize databases use them.

There have been Innumerable people who have helped me. Of course, I've managed to piss most of them off.
xam
Occasional Advisor

Re: What block size should be set ??

64K... I see.

is there any method to measuree the performance?

or only use silly method, start the time watch and copy a large file and then stop it?


ANy suggestion?

Thanks!