[Gluster-users] Slower than NFS

[Greg]

Hi,

I have a production setup, with clients that are both NFS and GlusterFS 
clients. I've made a _very_ quick benchmark (repeated ten times), to 
compare read performance between local hardrives, a NFS mount and a 
GlusterFS mount.
- a 17k file : 
cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae
- a 50M file : users0.new.spi
- /nfs is a NFS mounted point, server with SATA drives
- /gluster is a GlusterFS mounted point, 2x servers with SAS drives

php-14:~# for f in 
/var/www/www/cache/cache_c/cache_cf/cache_cfb/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae 
/nfs/tmp/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae 
/tmp/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae 
/gluster/sphinx/users0.new.spi /nfs/tmp/users0.new.spi 
/tmp/users0.new.spi; do echo $f; time cat $f >/dev/null; done
/var/www/www/cache/cache_c/cache_cf/cache_cfb/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae

real    0m0.011s
user    0m0.000s
sys     0m0.000s
/nfs/tmp/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae

real    0m0.002s
user    0m0.000s
sys     0m0.000s
/tmp/cache_c21f969b5f03d33d43e04f8f136e7682_c57d91c82f2ed2e96c13765e33fcd5ae

real    0m0.001s
user    0m0.000s
sys     0m0.000s
/gluster/sphinx/users0.new.spi

real    0m0.081s
user    0m0.000s
sys     0m0.068s
/nfs/tmp/users0.new.spi

real    0m0.037s
user    0m0.000s
sys     0m0.036s
/tmp/users0.new.spi

real    0m0.028s
user    0m0.000s
sys     0m0.028s


Configuration:
# file: /etc/glusterfs/glusterfsd.vol

#
# Volumes
#
volume media-small
        type storage/posix
        option directory /var/local/glusterfs/media_small
end-volume

volume media-medium
        type storage/posix
        option directory /var/local/glusterfs/media_medium
end-volume

# Lock posix
volume media-small-locks
  type features/posix-locks
  option mandatory-locks on
  subvolumes media-small
# subvolumes trash # enable this if you need trash can support (NOTE: 
not present in 1.3.0-pre5+ releases)
end-volume

volume media-medium-locks
  type features/posix-locks
  option mandatory-locks on
  subvolumes media-medium
# subvolumes trash # enable this if you need trash can support (NOTE: 
not present in 1.3.0-pre5+ releases)
end-volume


#
# Performance
#
volume media-small-iot
 type performance/io-threads
 subvolumes media-small-locks
 option thread-count 4 # default value is 1
end-volume

volume media-small-wb
 type performance/write-behind
 option flush-behind on # default is off
 subvolumes media-small-iot
end-volume

volume media-medium-iot
 type performance/io-threads
 subvolumes media-medium-locks
 option thread-count 4 # default value is 1
end-volume

volume media-medium-wb
 type performance/write-behind
 option flush-behind on # default is off
 subvolumes media-medium-iot
end-volume



#
# Serveur
#
volume server
        type protocol/server
        option transport-type tcp/server
        option auth.addr.media-small-wb.allow 10.0.*.*
        option auth.addr.media-medium-wb.allow 10.0.*.*
        # Autoconfiguration, e.g. :
        # glusterfs -l /tmp/glusterfs.log --server=filer-04 ./Cache
        option client-volume-filename /etc/glusterfs/glusterfs.vol
        subvolumes media-small-wb media-medium-wb # volumes exportés
end-volume

# file: /etc/glusterfs/glusterfs.vol


#
# Clients
#
volume media-small-filer-04
        type protocol/client
        option transport-type tcp/client
        option remote-host filer-04.local
        option remote-subvolume media-small-wb
end-volume

volume media-small-filer-05
        type protocol/client
        option transport-type tcp/client
        option remote-host filer-05.local
        option remote-subvolume media-small-wb
end-volume

volume media-medium-filer-04
        type protocol/client
        option transport-type tcp/client
        option remote-host filer-04.local
        option remote-subvolume media-medium-wb
end-volume

volume media-medium-filer-05
        type protocol/client
        option transport-type tcp/client
        option remote-host filer-05.local
        option remote-subvolume media-medium-wb
end-volume


#
# Volume principal
#
volume afr-small
        # AFR has been renamed to "Replicate" for simplicity.
        type cluster/replicate
        # Il faut mettre le serveur avec le moins d'espace disque en 1er :
        # "When doing a "df -h" on a client, the AVAILABLE disk space 
will display the maximum disk space of the first AFR sub volume defined 
in the spec file. So if you have two servers with 50 gigs and 100 gigs 
of free disk space, and the server with 100 gigs is listed first, then 
you will see 100 gigs available even though one server only has 50 gigs 
free. "
        subvolumes media-small-filer-04 media-small-filer-05
end-volume

volume afr-medium
        # AFR has been renamed to "Replicate" for simplicity.
        type cluster/replicate
        subvolumes media-medium-filer-04 media-medium-filer-05
end-volume

#
# Performance
#
volume iot-small
        type performance/io-threads
        subvolumes afr-small
end-volume

volume readahead-small
        type performance/read-ahead
        subvolumes iot-small
        option page-size 1MB         # default is 256KB
        option page-count 4          # default is 2 - cache per file = 
(page-count x page-size)
        option force-atime-update no # defalut is 'no'
end-volume

volume iocache-small
        type performance/io-cache
        option cache-size 64MB          # default is 32MB
        option page-size 256KB          # default is 128KB
        subvolumes readahead-small
end-volume

volume writeback-small
        type performance/write-behind
        option block-size 1MB # max 4MB
        option flush-behind on # default is off
        subvolumes iocache-small
end-volume

volume iot-medium
        type performance/io-threads
        subvolumes afr-medium
end-volume

volume readahead-medium
        type performance/read-ahead
        subvolumes iot-medium
        option page-size 1MB         # default is 256KB
        option page-count 4          # default is 2 - cache per file = 
(page-count x page-size)
        option force-atime-update no # defalut is 'no'
end-volume

volume iocache-medium
        type performance/io-cache
        option cache-size 64MB          # default is 32MB
        option page-size 256KB          # default is 128KB
        option page-count 2
        subvolumes readahead-medium
end-volume

volume writeback-medium
        type performance/write-behind
        option block-size 1MB # max 4MB
        option flush-behind on # default is off
        subvolumes iocache-medium
end-volume

Regards,

-- 
Greg