[Gluster-devel] how to set the default read/write block size for all transactions for optimal performance (e.g. anything similar to rsize, wsize nfs options?)

[Sabuj Pattanayek]

Hi,

On Sun, Jul 20, 2008 at 11:51 AM, Anand Avati <avati at zresearch.com> wrote:
> write-behind is not being used in  your configuration. You need to chain the
> performance translators.

After chaining the translators:

#### glusterfs-server.vol ####

volume eon0
  type storage/posix
  option directory /export/eon0
end-volume

volume iothreads-eon0
  type performance/io-threads
  #option thread-count 4  # deault is 1
  option cache-size 64MB
  subvolumes eon0
end-volume

volume writebehind-eon0
  type performance/write-behind
  #option aggregate-size 131072 # in bytes
  option aggregate-size 1MB # default is 0bytes
  option flush-behind on    # default is 'off'
  subvolumes iothreads-eon0
end-volume

volume readahead-eon0
  type performance/read-ahead
  # enabled on server and client 57MB/s -> 61MB/s
  option page-size 65536 ### in bytes
  option page-count 16 ### memory cache size is page-count x page-size per file
  subvolumes writebehind-eon0
end-volume

volume server
  type protocol/server
  option transport-type tcp/server
  option auth.ip.eon0.allow 10.2.179.*
  subvolumes eon0
end-volume

####

#### glusterfs-client.vol ####
volume eon0
  type protocol/client
  option transport-type tcp/client
  option remote-host porpoise-san
  option remote-subvolume eon0
end-volume

volume iothreads-eon0
  type performance/io-threads
  #option thread-count 4  # deault is 1
  option cache-size 64MB
  subvolumes eon0
end-volume

volume writebehind-eon0
  type performance/write-behind
  # huge boost, 12MB/s -> 50MB/s
  #option aggregate-size 131072 # in bytes
  # 50MB/s -> 54MB/s
  option aggregate-size 1MB # default is 0bytes
  # 54MB/s -> 57MB/s
  option flush-behind on    # default is 'off'
  subvolumes iothreads-eon0
end-volume

volume readahead-eon0
  type performance/read-ahead
  option page-size 65536 ### in bytes
  #option page-size 131072 ### in bytes
  option page-count 16 ### memory cache size is page-count x page-size per file
  subvolumes writebehind-eon0
end-volume

volume io-cache-eon0
  type performance/io-cache
  # Doesn't really help
  option cache-size 64MB             # default is 32MB
  # This gives a little bit of boost 9MB/s to 12MB/s
  option page-size 1MB               #128KB is default option
  #option priority *.h:3,*.html:2,*:1 # default is '*:0'
  option priority *:0
  # 500 KB/s boost
  option force-revalidate-timeout 2  # default is 1
  subvolumes readahead-eon0
end-volume

####

I'm getting 60-62MB/s with Gluster, When I use NFS without specifying
rsize or wsize I also get 60MB/s. When I use the NFS mount options
rsize=262144,wsize=262144 and do a dd with the "standard" bs=4k I get
80MB/s (increasing bs= no longer helps for NFS at this point) . With
Gluster I can get 100-110MB/s if I use bs=128k (which is faster than
GFS2!). So once again the main question here is there something for
Gluster like the rsize or wsize options or is there something else I
can tweak in the translators that would be the equivalent of
increasing the global rsize and wsize? If not is there something I can
patch in Fuse or in the Gluster sources to do this?

Thanks,
Sabuj