[Gluster-users] cluster/distribute vs DHT

[Krishna Srinivas]

Hi Frederico,

You guessed it right, distribute is the new name for DHT as the name
"distribute" is more intuitive.

Here is the volfile for the client with distribute+replicate with out
any performance translator:
-----------
volume client1
type protocol/client
option transport-type tcp
option remote-host sev1
option remote-subvolume brick
end-subvolume

volume client2
type protocol/client
option transport-type tcp
option remote-host sev2
option remote-subvolume brick
end-subvolume

volume client3
type protocol/client
option transport-type tcp
option remote-host sev3
option remote-subvolume brick
end-subvolume

volume client4
type protocol/client
option transport-type tcp
option remote-host sev4
option remote-subvolume brick
end-subvolume

volume client5
type protocol/client
option transport-type tcp
option remote-host sev5
option remote-subvolume brick
end-subvolume

volume client6
type protocol/client
option transport-type tcp
option remote-host sev6
option remote-subvolume brick
end-subvolume

volume replicate1
type cluster/replicate
subvolumes client1 client2
end-volume

volume replicate2
type cluster/replicate
subvolumes client3 client4
end-voloume

volume replicate4
type cluster/replicate
subvolumes client5 client6
end-volume

volume distribute
type cluster/distribute
subvolumes replicate1 replicate2 replicate3
end-volume
-------------

If you need performance translators, you can add these translators to
the end of the file:
---------

## Add readahead feature
volume readahead
type performance/read-ahead
option page-size 1MB # unit in bytes
option page-count 2 # cache per file = (page-count x page-size)
subvolumes distribute
end-volume

## Add IO-Cache feature
volume iocache
type performance/io-cache
option page-size 256KB
option page-count 2
subvolumes readahead
end-volume

## Add writeback feature
volume writeback
type performance/write-behind
option aggregate-size 1MB
option window-size 2MB
option flush-behind off
subvolumes iocache
end-volume
----------------------

For replicate (for nufa like effect), we have an option "option
read-volume <volname>" which ensures that replicate always does the
reads from the volume configured (which is usually the local
storage/posix volume)

Let us know if you have any more questions.

Thanks
Krishna


On Thu, May 14, 2009 at 7:42 AM, Sacerdoti, Federico
<Federico.Sacerdoti at deshawresearch.com> wrote:
> Hi,
>
> I am benchmarking gluster, and would like to ensure I understand the options
> related to consolidating the storage bricks. I'm a bit confused about
> "cluster/unify", "cluster/dht", and "cluster/distribute".
>
> This page shows how dht is different from unify, but what is distribute, a
> rename of DHT?
>
> http://www.gluster.org/docs/index.php/Understanding_DHT_Translator
>
> It would be helpful to see an example of distribute+replicate. I am trying
> to approximate an HDFS setup that I've tested, which replicates files over
> many servers, but favors local storage like nufa. In Gluster I am using:
>
> Translators:
>  fuse -> distribute -> replicate -> {bricks}
> Bricks are:
>  server -> io-threads(8) -> locks -> posix_storage
>
> I would like to measure worst-case performance, without any cache effects.
> Are there any translators or options that I should include for best
> performance?
>
> Thanks,
> fds
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://zresearch.com/cgi-bin/mailman/listinfo/gluster-users
>
>