How would data be spread on a meta volume

Question

If a server needs 4 drives (16 GB, 40 GB, 40 GB and 104 GB), is it better to create 1 200 GB meta and divide it logically on the server or 4 metas of the appropriate sizes?

dynamox · Accepted Answer

My main concern is that if Ipresented 1 meta to the server and a logical volumeof 16 GB was created, that it would be spread acrossthe many hypers and not allocated on the first 2 8 GB hypers. If you create a striped meta ..then it will be spread out on 25 hypers. I don't know what OS you are using for Oracle ..but also think about future expansion. If you are running Oracle on some flavor of Unix ..it would be much easier to expand in the future by utilizing some sort of volume manager (veritas, LVM ..etc). I would definitely not mix Oracle Database and Application on the same meta, but i would actually take it a step further and use different meta's for Oracle. I would create separate meta's for my logs, separate metas for my database. Access patterns are so different, if you have 2-way mirrors ..i would use those for logs ..and maybe RAID-5 for database. Let's say if you run Oracle on windows ...if you use striped meta and have to expand it ...big pain in the a$$, it involves saving data to BCV first, then adding another member to your meta which restripes all the data and then restore data from BCV. Of course you could use concatenated meta for easy of expansion ..but there goes your performance.

MrTS2Symm · Answer

Danne,

Dynamox is correct.

For performance and data layout, you would want to spread out your workload. It sounds like from your question, there will be 4 different applications or uses for the 4 drives and this supports your original needs.

dynamox · Answer

4 metas is better ...you will spread it on more spindles, thus giving you more IOPS, queues ..etc

Danne-Murphy · Answer

It is an application and an Oracle database. I understand about the 4 metas having addition IO queues on the host, but a 200 GB meta would have 25 8.4 GB hypers and if I was careful in creating the 4 metas, they would be spread over the same 25 hypers. So, wouldn't the io load across the drives be the same in that case. My main concern is that if I presented 1 meta to the server and a logical volume of 16 GB was created, that it would be spread across the many hypers and not allocated on the first 2 8 GB hypers.

dynamox · Answer

if you have only one big meta ..you will have only one disk queue ..because powerpath can only talk to the meta head ...it can't talk to meta members. So you will limit yourself there.

Danne-Murphy · Answer

Thanks for the information. Your thoughts will be taken into account and we will do some redesign on how the disks are configured.

dynamox · Answer

If you create a striped meta ..then it will be spread out on 25 hypersi thought i said hypers

sysmgr1 · Answer

I disagree with dynamox in that it should use the same number of spindles whether you make one big meta or 4 smaller ones that equal the same size. You use the same number of hypers either way.

I do agree with 4 smaller metas is better though for the reasons everyone listed. Also, you potentially can gain additional cache areas along the IO path in various places. (ie. filesystem cache.)

dynamox · Answer

I disagree with dynamox in that it should use the
same number of spindles whether you make one big meta
or 4 smaller ones that equal the same size. You use
the same number of hypers either way.

you bring up a good point ..if you use 25 unique hypers, that does not quarantee that you will end up on 25 unique physical spindles. If we take a 146G drive and slice it into ~8.6G hypers ..that will be a about 16 hypers per physical disk. So it's going wrap around at some point and you might select hypes for your meta on the same disk. When my CE builds the bin file, Symmwin spreads hypers as random as possible. So when i run symdev list, hypers are listed in very random way, different DAs, different interfaces. If i am building a 34G meta, i just pick the first 4 8.6G hypers. Knowing that everything in DMX goes thru cache and there is no direct read/write to disk ..i am not as picky building my meta, Clariion would be a different story of course.

Symmetrix

How would data be spread on a meta volume

Was this post helpful?