Re: RAID Question

From: benlag@aent.com > >I know that this subject has been very popular in this forum but I >still have questions concerning my specific scenario. > >I am moving all my data from my Informix 7.31 FC6-1 Database that >resides on an HP KClass box to an SAG disk array box. The database is a >data warehouse and is optimized for DSS. The total disk space amounts >to 1 terabyte of which 500GB will be machine mirrored. There are 16 >disks which are approximately 36MB each that can be used for storage. > >I have researched and read much information on RAID levels in this >forum and on the internet and believe RAID 10 is my best option. My >only concern is the what my Unix Admin is suggesting by setting up the >striping using the OS or should I let Informix do it through >fragmentation. My Unix Admin is suggesting using 2 disks (each 36GB) No reason you can't do both. >to create each raw device(volume group). This leaves me with only 8 >volume groups (each of them 72Gig in size). As a result this leaves me >with just 8 physical disks(groups) to work with in Informix. > >I am arguing about doing the striping myself using Informix >fragmentation to increase flexibility and administration of the disks. >Here are example scenarios to make this less confusing: > >Example1 (Unix admin setup (striping done at os level)): >8 volume groups (each 72Gig in size): >vg01 vg02 vg03 vg04 vg05 vg06 vg07 vg08 > >Each logical volume in that group is 2Gig in size and each group >consists of 36 logical volumes: >lv01,lv02,lv03,lv04,lv05.....lv36 for each group > >Each dbspace will contain 4 logical volumes all of which are in the >same group (this gives me max of 72 dbspaces): >dbsp1:/vg01/lv01,/vg01/lv02,/vg01/lv03,/vg01/lv04 >dbsp2:/vg02/lv01,/vg02/lv02,/vg02/lv03,/lvg02/lv04 >dbsp3,dbsp4.... dbsp*** and so on > >I create a database and fragment my tables across 4 of those volume >groups and my indexes detached: > >create mytable .... fragment by round robin in dbsp1,dbsp2,dbsp3,dbsp4; >create index idx_mtbl in dbsp5>create my2ndtable .... fragment by round robin in >dbsp5,dbsp6,dbsp7,dbsp8; >create index idx_mtbl in dbsp1> >Let's just say I run this query: >select mytable.field1,my2ndtable.field2 >from mytable,my2ndtable >where mytable.linkfield=my2ndtable.linkfield >and mytable.idxfield="1"> >Now, knowing that the indexes for each table are in the same disks as >the linking table would this not defeat the purpose of fragmenting. Is >the above disk config and setup better than this one: > >Example2 (Striping done by Informix): > >16 volume groups (each 36 GB) >vg01 vg02 vg03 vg04 vg05 vg06 vg07 vg08...vg16 > >Each logical volume in that group is 2Gig in size and each group >consists of 18 logical volumes: >lv01,lv02,lv03,lv04,lv05.....lv18 for each group > >Dbspaces the same size as above (4 logical volumes) but 16 disks allows >me to create 4 sets of 4 dbspaces. >dbsp1,dbsp2,dbsp3,dbsp4....dbsp** > >create mytable .... fragment by round robin in dbsp1,dbsp2,dbsp3,dbsp4; >create index idx_mtbl in dbsp5>create my2ndtable .... fragment by round robin in >dbsp6,dbsp7,dbsp8,dbsp9; >create index idx_mtbl in dbsp10> > >Wouldn't fragmentation and the data being spread out over 16 disks be a >better solution because of the fact that I will have more disk heads >reading at the same time with each query? Or should I take the Unix >admin's advice and let the OS handle the striping?? > > >Any help or suggestions would be greatly appreciated, > > >Benny Lago >Alliance Entertainment, Inc. > > > > > >Sent via Deja.com http://www.deja.com/ >Before you buy. _____________________________________________________________________________________ Get more from the Web. FREE MSN Explorer download : http://explorer.msn.com