"Oracle archive files and redo log files have a predictable workload composed mainly of sequential writes. These can be efficiently handled by the array’s write cache and assigned HDDs.

Oracle database on EMC VNX Best Practices

http://powerlink.emc.com/km/live1/en_US/Offering_Technical/White_Paper/h8242-deploying-oracle-vnx-wp.pdf

Thanks,

Eddy

R10 is recommended for Redo Log and R5/R6 is recommend for Achieve log files.

R10 has less write penalty than R5, therefore it is better for Oracle redo log, seq write IO, and the configuration of R10 with Storage SP Cache could manage redo log in most of situation very well.

Thanks,

Eddy

Are you familiar what full stripe write is ?

Dynamox,

You can review this document that discus Raid 5 and Raid 6 full stripe writes

I hope this helps you

http://www.emc.com/collateral/hardware/white-papers/h2891-clariion-raid-6.pdf

I know what full stripe write is, i am trying to argue my point that due to write pattern and write size raid 5 full stripe write could be just as efficient and have less write penalty 1.25 compared to 2 for raid 10.

Sent from my Verizon Wireless 4GLTE Phone

Write penalty of R5 is 4, i.e. with R5(4+1), it is twice than that of R10, which is 2. Online Redo log is best with R10 and avoid R5.

Here is an article for your reference.

http://www.dba-oracle.com/t_online_redo_logs_raid.htm

Thanks,

Eddy

do some research on what MR3 Write is

Well, MR3 write require LBA of data in SP cache is continuous. It happen when host IO size is multiplex of RAID strip size.

In Oracle environment, Storage cache is filled with mixture of DBWR data and LGWR data, therefor MR3 full strip write happens little.

Thoughts?

Eddy

interesting !

Not very sure what does "SP does not care, if I/O fills the RAID stripe MR3 will be used to write that data to the disks." mean by dynamox. MR3 only occurs when a continuous full stirpe data can be generated in the write cache through any possible mechanism, say "Back Fill" for CLARiiON.

Dynamox,

Could you elebrate your point of “SP does not care…”?

The following is my understanding.

MR3 is used by OE/FLARE to reduce WP of R5 when IO cached in SP is with continuous address belong to a RG.

In ORA environment, SP cache would take IO both of DBWR or LGWR, one is Random other is sequential, before IO of LGWR filled slots of MR3, the page of Cache might be flushed already, therefor I do not think MR3 happens on LGWR IO.

Thoughts?

Eddy

So you are saying sp cache can only handle only reads or writes not both..we know thats not true.

Umm..isnt that what i said ..stripe is full and you mr3 write.

no, Eddy is saying diffrent(at least with respect to MR3); Dynamox's point is R5 has advantag over R10 with 1.25 right penalty ( with full strip WR and MR3) , but Eddy is "thinking" MR3 does not happen on LGWR IO

It sounds  to me(based on what both of u debating( percentage of MR3 would be lesser, but not like there wont be any MR3 at all.