EVA and Oracle Databases.

[DeafFrog]

As i understand , EVA have inbuild "V-Raid" ,would like to know
-if it's a OK to assign a single big LUN , say of 280G ,create a single Vg , and create LVs to contain all Database FS ( indexes , redolog , archive ,datafiles ..etc).We are using Vraid 1 , and EVA 8K is configured in 2C2D configuration.Some ppl. will suggest that in vraid striping and mirroring are taken care of and we need not worry about db block size etc .
-could you please share you experience with database(Oracle on EVA).
-we are expecting good number of hits to Database.

[IBaltay]

Hi,
the general best practices is using the EVA LUNs/LVM as follows:

redo logs RAID1, each as a separate EVA LUN
archive logs RAID1, each as a separate EVA LUN
redo log/archive logs mirrors on a separate EVA Luns
datafiles RAID5 nicely balanced according to the number of the front ports/host hbas, etc.,
e.g. 4 smaller LUNs, 8 smaller LUNs for LVM,...

Also it is a good practice in HPUX before 11.31 to evenly set preferred path to controller a,b (failover, failback) on the eva controllers during the vdisk presentation

[DeafFrog]

Thankyou IBaltay
hp os is 11.31 , setting up multipathing manually is ruled out,Oracle version will be
10.2.0.7.expecting more views!! thanks once agian

[Uwe Zessin]

> setting up multipathing manually is ruled out

I don't know if 11.31 understands ALUA (asymmetrical logical unit access) - if not, you better DO manual optimizations or all virtual disks can end up on one EVA controller over time.

Oh, I just see you have a 2C2D - well, it should not be that bad on such a small box...

[emha_1]

"the general best practices is using the EVA LUNs/LVM as follows"

@IBaltay
is it best practise according ORACLE or HP?

I have configured LUNs for my databases almost just the same way (just archived redologs I've put on R5 - log switch once in 20min or so can R5 handle pretty fine, I think) but unfortunately time to time I'm reaching throughput capacity limits on EVA8k controler's internal mirror ports. storage vendor (after discussion with HP) tries to persuade me to change R5 to R10 because R10, allegedly, should generate less internal traffic.

I do not incline to such a change due to several reasons.
Do you have a link to any officially released best practice document (preferably from HP) according topic? If you could share it I would appreciate it.

thanks.

emha.

[IBaltay · Accepted solution]

Hi,
... but unfortunately time to time I'm reaching throughput capacity limits on EVA8k controler's internal mirror ports. storage vendor (after discussion with HP) tries to persuade me to change R5 to R10 because R10, allegedly, should generate less internal traffic.

EVA general perf best practices:
ftp://ftp.compaq.com/pub/products/storageworks/whitepapers/5982-9140EN.pdf

HPUX 11.31 Multipathing:
HPUX11.31 - MPIO Native (ALUA/ALBA)
http://docs.hp.com/en/native-multi-pathing/native_multipathing_wp_AR0803.pdf

Evaperf for the VDISK perf analysis:
http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1239472

RESUME/Notes
a) LVM on UX/Linux to nicely loadbalance the front end
b) the more physical disk in the DG the more backend distribution of each VDISK
c) Multipathin solution, either manual (PVlinks) or automatic (Native MPIO)

Few words on EVA Active/Active and loadbalancing:
a) all writes are being written into both controllers via the mirror ports

b) all reads are driven directly via the managing controller, undirectly via the mirror port

Host MPIO ALUA optimizes this and communicates directly with EVA managing controller which then reduces the mirror port based reads by the second (proxy) controller

[DeafFrog]

Thank you all Gentel Men .