[Gluster-users] Re :Performance of GlusterFs and NFS

[mohan L]

On Mon, Jan 12, 2009 at 6:57 PM, Keith Freedman <freedman at freeformit.com>wrote:

> check back to some of the previous messages in this list about performance.

Ok , I will check too .


>
> also, which version of Gluster you use makes a difference.

I am using

GlusterFS version: glusterfs 1.4.0rc7

FUSE Version     :fuse-2.7.3glfs10

My Client and Server machine  RAM size  : 384MB(128MB+256MB)

This the volume file I am using for benchmarking.


### file: client-volume.vol.sample

### Add client feature and attach to remote subvolume

volume client
  type protocol/client
  option transport-type tcp
  option remote-host 192.168.1.167
  option remote-subvolume brick
end-volume

### Add readahead feature
volume readahead
  type performance/read-ahead
  option page-size 128kB
  option page-count 4
  subvolumes client
end-volume

### Add IO-Cache feature
volume iocache
  type performance/io-cache
  subvolumes readahead
  #option page-size 1MB      # 128KB is default
  option cache-size 300MB    # 32MB is default
  option cache-timeout 5 # 1second is default
  #option priority *.html:2,*:1 # default is *:0
end-volume

### Add writeback feature
volume writeback
  type performance/write-behind
  option aggregate-size 128KB
  option window-size 2MB
  option flush-behind on
  subvolumes iocache
end-volume




### file: server-volume.vol.sample

volume brick
  type storage/posix
  option directory /bench
end-volume

volume plocks
  type features/posix-locks
  subvolumes brick
 # option mandatory on
end-volume

volume iot
  type performance/io-threads
  subvolumes plocks
  option thread-count 8
 # option cache-size 4MB
end-volume

volume server
  type protocol/server
  subvolumes iot brick
  option transport-type tcp
  option auth.addr.brick.allow  192.168. *
  option auth.addr.iot.allow 192.168.*
end-volume

I am using  io-threads on server side only .can I use same on client side?
It will increase the performance ? How to tune translator parameters? I mean
I am using page-size in iocache 128K how to know which parameter will
performs better ?

Next I am aiming to set up more server and client to get the performance
gain .

we are aiming at solving the problems of Millions/Billions of small files
inside a single directory, Though, the files are not saved as files inside
each directory, they will be saved inside a db file .We are using MySQL .
How to achive this for testing perpose? As know I don't have enough skill to
do same . Please give me some guide line to achive this kind of set up . I
am not able to get how to export db file over GlusterFS. what is your Idea?





>
>
> As for smaller block sizes being a performance issue, I think mostly fuse
> is the problem with small block sizes.
> I think 4K is what fuse' block size is?   SO I'm thinking that should be ok
> for testing.




>
>
> I also believe gluster 2.0 is much faster with smaller files than 1.3, so
> you should get better results with 2.0


I saw last week one thread about 2.0 .But I am not able to find  where I
have to download 2.0? Please give me the download URL of 2.0 .


>
>
> if you benchmarked with 1.3, please try again with 2.0 and let us know your
> results.

 Sure , I will send 2.0 benchmark also .

Thanks for your time

Thanks
L.Mohan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://supercolony.gluster.org/pipermail/gluster-users/attachments/20090113/3481369f/attachment.html>