Re: VLDB ASM & SAN Striping Question
Date: Thu, 7 Jun 2012 15:52:19 +0300
Message-ID: <CAKA5Cb+eyDY-kEop=soYtUNNDHf7-Fur9rfAQS3S3J=sqsKL7w_at_mail.gmail.com>
Hi Chris,
Just to share my experience, sorry, but it's particularly for EVA as I
have most experience with. For several years now I've never
experienced performances issues with EVA and I'm very happy so far.
Yes, HP recommends putting all the drives in one disk group, by doing
so you're supposed to get best performance, which will scale if you
add more disk drives. Yes, creating a disk group in EVA will span
across all disk drives and LUNs in this disk group will span across
all disk drives within the disk group. The only few times I saw disks
divided into different disk groups was because of different disk kind
(FC, FATA, SSD) and/or FC disks with different rotation speed (10k,
15k). Just to know - disk group redundancy has two options - single
protection or double protection. Meaning that you won't loose any data
in case of single disk failure or for double protection failure of two
disks failures. There are no spares here, but the space for protection
is reserved within the disk group.
Now, there are two problems with EVA - controller cache is one for all disk groups i.e. disk group doing more operations will occupy more of the cache space. Second there is no I/O prioritization, neither per disk group, nor per LUN.
A problem I had with EVA4400 was that customer had two disk groups - one with 30 FC drives and one with 50 FATA drives. What was happening was that LUNs on the FATA drives saturated the cache of the controller. As an immediate result we saw that log_file_sync of database running on the FC disks increased by times of 10x or more and this wasn't even in the business hours. The only solution for this was to separate the LUNs of FATA drives on one controller and LUNs of FC drives to the other one.
So particularity for EVA I don't see any real benefit of spiting disks
(one of a kind) in more than one disk group. Again, it depends,
without I/O prioritization one could prefer to separate critical
database in different disk group. The only configuration you could
take care of is to divide LUNs equally on both controllers (which EVA
is doing by default).
Usually I create at least two LUNs with proper size in VRAID5 and build on top of them ASM disk group with external redundancy.
Here are few good papers for reference:
http://h71028.www7.hp.com/enterprise/downloads/4AA0-9728ENW_ASM_EVA_Best_Practices_White_Paper_final%20121806.pdf http://h20195.www2.hp.com/v2/GetPDF.aspx/4AA1-6194ENW.pdf http://h20195.www2.hp.com/v2/GetPDF.aspx/4AA1-5658ENW.pdf
Regards,
Sve
-- http://www.freelists.org/webpage/oracle-lReceived on Thu Jun 07 2012 - 07:52:19 CDT