[Gluster-users] 2.0.8: server on tmpfs for small files

Jeremy Enos jenos at ncsa.uiuc.edu
Sat Nov 20 02:11:28 UTC 2010


For an experiment, I tried dumping a loopback filesystem into the tmpfs 
filesystem, and then exporting that one w/ Gluster.  It seems to have 
worked for any single client, though I'm not getting IOR to work against 
it yet w/ MPI.  I'm expecting performance to stink, but we'll see.

     Jeremy

On 11/19/2010 7:12 PM, Jeremy Enos wrote:
> Looks like RAMFS has the same issue TMPFS does...  I'm looking into 
> the RNA networks.  Thanks-
>
>     Jeremy
>
> On 11/18/2010 6:55 PM, Craig Carl wrote:
>> On 11/18/2010 04:33 PM, Jeremy Enos wrote:
>>> Post is almost a year old...  ever any response here?  Is it 
>>> possible to export tmpfs locations w/ gluster?
>>> thx-
>>>
>>>     Jeremy
>>>
>>> On 12/1/2009 8:14 AM, Alexander Beregalov wrote:
>>>> Hi
>>>>
>>>> Is it possible to start server on tmpfs ?
>>>> It is announced that stripe can be used over tmpfs, but stripe is
>>>> client plugin and server cannot start on tmpfs because lack of
>>>> xattrs.
>>>>
>>>> I am trying to setup a small fast storage for small files 
>>>> (compiling purpose).
>>>> I made ext2 with xattr on ramdisk on 4 hosts, joined them with
>>>> replicate plugin and mounted it on one client. Also io-cache,
>>>> write-behind, quick-read and io threads  were used on client side.
>>>> I compiled linux kernel, performance was 10 times worse than tmpfs
>>>> exported by NFS on one node.
>>>>
>>>> Any ideas?
>>>>
>>>> Servers:
>>>>
>>>> volume posix
>>>>    type storage/posix                   # POSIX FS translator
>>>>    option directory /mnt/ost        # Export this directory
>>>> end-volume
>>>>
>>>> volume locks
>>>>    type features/locks
>>>>    option mandatory-locks on
>>>>    subvolumes posix
>>>> end-volume
>>>>
>>>> volume brick
>>>>    type performance/io-threads
>>>>    option thread-count 4                # Four CPUs
>>>>    subvolumes locks
>>>> end-volume
>>>>
>>>> volume server
>>>>    type protocol/server
>>>>    option transport-type tcp
>>>>    option transport.socket.nodelay on
>>>>    subvolumes brick
>>>>    option auth.addr.brick.allow * # Allow access to "brick" volume
>>>> end-volume
>>>>
>>>>
>>>> Client:
>>>>
>>>> volume server1
>>>>    type protocol/client
>>>>    option transport-type tcp
>>>>    option remote-host<IP>
>>>>    option transport.socket.nodelay on
>>>>    option remote-subvolume brick        # name of the remote volume
>>>> end-volume
>>>>
>>>> <the same for server[2-4]>
>>>>
>>>> volume replicated
>>>>    type cluster/replicate
>>>>    subvolumes server1 server2 server3 server4
>>>> end-volume
>>>>
>>>> volume iocache
>>>>    type performance/io-cache
>>>>    option cache-size 1000MB        # default is 32MB
>>>>    option priority *.h:3,*.o:2,*:1     # default is '*:0'
>>>>    option cache-timeout 1                 # default is 1 second
>>>>    subvolumes replicated
>>>> end-volume
>>>>
>>>> volume writeback
>>>>    type performance/write-behind
>>>>    option cache-size 500MB     # default is equal to aggregate-size
>>>>    option flush-behind off      # default is 'off'
>>>>    subvolumes iocache
>>>> end-volume
>>>>
>>>> volume quickread
>>>>    type performance/quick-read
>>>>    option cache-timeout 1         # default 1 second
>>>>    option max-file-size 256KB        # default 64Kb
>>>>    subvolumes iocache
>>>> end-volume
>>>>
>>>> volume iothreads
>>>>    type performance/io-threads
>>>>    option thread-count 16 # default is 16
>>>>    subvolumes quickread
>>>> end-volume
>>>> _______________________________________________
>>>> Gluster-users mailing list
>>>> Gluster-users at gluster.org
>>>> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>>>>
>>> _______________________________________________
>>> Gluster-users mailing list
>>> Gluster-users at gluster.org
>>> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>>
>> Jeremy -
>>    xattrs are required for any Gluster setup regardless of the volume 
>> design, tmpfs doesn't support user xattrs. RamFS works well, if you 
>> have the budget FisionIO cards are very fast and work well with 
>> Gluster, so does the solution from RNA Networks. 
>> (http://www.rnanetworks.com/)
>>
>> Thanks,
>>
>> Craig
>>
>> -->
>> Craig Carl
>> Senior Systems Engineer
>> Gluster
>> _______________________________________________
>> Gluster-users mailing list
>> Gluster-users at gluster.org
>> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>



More information about the Gluster-users mailing list