[Gluster-devel] Questions on ALU/unify

DeeDee Park deedee6905 at hotmail.com
Wed Aug 1 09:11:52 UTC 2007

Here is the testing I was doing. I got different results than expected.

I have 2 bricks (2 Hard drives), 1 client on the same machine, and another

source data disk for testing.
The two bricks are 500GB each and source disk is 1TB.

I filled up one 500GB drive with about 450GB of data, and the other 500GB

drive was empty.
I unify the two bricks, and cleared the namespace cache, and let it build a

new namespace cache.

I then rsync my 1TB, about 660K files into the unify client.

what I expected was that all new files that weren't already in the full

brick, would store it into the empty brick because the empty brick had more

resources available. What I saw was that files from my full brick were


Test is still going on, but currently I see about 150GB on the empty brick,

and about 350GB on the full brick.

What I suspect happened was that the ALU (or unify) placed the copy of the

file equally between the two, removing it from the full brick -- effectively

redistributing the files according to the ALU algorithm. If the files

already existed on a brick, I assumed the ALU (or unify or whomever is doing

it), would not move the files. this is great if I wanted to redistribute the

files, but it was at a cost of performance. It seems that when files are
removed, they are copied to the other brick, thus chewing up extra 


More information about the Gluster-devel mailing list