I have a very write intensive file system that is 600 GB and does about 1,500 IOPS on an EMC cx380 array. I have created a Raid 1/0 Raid Group using 16 disks 8+8. I have created six 100 GB luns.
The statistical application used on this server typically writes, sorts, merges and manipulates data on this file system. With 20 users working the file system will be inserting and deleting data on the first 100 GB (one lun) of the file system. As more users run jobs the file system work space may go to 300GB or 400GB (3 or 4 luns). Finally when all users are logged in and running jobs it may use all 6 luns.
When there is lower usage it will only use one lun but I want it to use all six at all times. So when creating the logical volume for the file system in smit I specified striping with a 64k stripe size. Now when the application writes six blocks of data it should stripe accross all six luns in a balanced fashion (I hope). I realize I can use meta luns but want to hold off until I need to expand the file system in the future. At that time I may stripe expand the six lun to another six luns in a future 8+8 raid group.
I'm not sure if I picked the correct stripe size. I read somewhere that the stripe size in smit is actually the stip not stripe size. That is my strip size is 64k but my stripe size (64k x 6 luns) is really 384k.
The cx-380 array has an element size of 128 blocks. Each block is 512 bytes so each write or stripe will be 64k. See link for info -
Based on this did I pick the best stripe size in smit?
The statistical application used on this server typically writes, sorts, merges and manipulates data on this file system. With 20 users working the file system will be inserting and deleting data on the first 100 GB (one lun) of the file system. As more users run jobs the file system work space may go to 300GB or 400GB (3 or 4 luns). Finally when all users are logged in and running jobs it may use all 6 luns.
When there is lower usage it will only use one lun but I want it to use all six at all times. So when creating the logical volume for the file system in smit I specified striping with a 64k stripe size. Now when the application writes six blocks of data it should stripe accross all six luns in a balanced fashion (I hope). I realize I can use meta luns but want to hold off until I need to expand the file system in the future. At that time I may stripe expand the six lun to another six luns in a future 8+8 raid group.
I'm not sure if I picked the correct stripe size. I read somewhere that the stripe size in smit is actually the stip not stripe size. That is my strip size is 64k but my stripe size (64k x 6 luns) is really 384k.
The cx-380 array has an element size of 128 blocks. Each block is 512 bytes so each write or stripe will be 64k. See link for info -
Based on this did I pick the best stripe size in smit?