logs archiveIRC Archive / Freenode / #oracle / 2010 / April / 16 / 1
MxxCon
we will run asm on oracle data partition, which is fc san
but os and oracle itself will be sitting on local disk..does it require asm too?
TheBonsai
no. and no special structure
it's up to you and your needs
like
own filesystem for ora software or so
or a shared software installation (on NFS or cluster filesystem)
MxxCon
but is there some doc that shows best practices..like does it make sense to put ora software on its own partition and how big should it be? does it make sense to make /tmp separate, etc etc
i don't want to be in a position where a few month from now we'll find out that some partition is too small or too large
DiscordianUK
There's an Oracle document on recommended pratice
*practice
Only you know how big your dataset is and can project it's growth
MxxCon
i'm talking about only os + oracle software itself. not data
for data we have a separate san
DiscordianUK
oel=oracle unbreakable linux?
MxxCon
yes
DiscordianUK
The XE rpm is about 300 Mb I think
so a couple of gig installed
MxxCon
which is pretty much rhel..we picked oel so that there would be less fingerpointing in case of some problem :)
let oracle fight it out among themselves if its os or db issue :)
         

DiscordianUK
(Action) laughs
TheBonsai
once i had some class @oracle germany, they didn't even remove the red hat icon from their unbreakable desktop workstatsions...
DiscordianUK
heh
MxxCon
DiscordianUK do you know url to that recommended practice doc?
DiscordianUK
CentOS at least goes to a lot of trouble to rebrand
http://appsdba.com/dbconfig_file_layout.htm
Dormir
for the oracle software install, I would recommend local installs vs shared
(for RAC)
TheBonsai
yea, same here. local installs save trouble, just in case
MxxCon
yes, both servers will be using local raid1 disk for os and oracle software.
TheBonsai
personally i put ~oracle in /opt on a separate filesystem
might make sense - or not
(~oracle contains the installation, in this case, of course)
MxxCon
how big
TheBonsai
several gig, i don't remember
also depends on the edititon
how many separate installations do you want to have?
MxxCon
i have 500GB local disk
2 servers
TheBonsai
well, should be enough for a OS, an installation and some logfiles
you could use the remaining 480GB for your pr0n
seriously, how many ORACLE_HOMEs do you want to roll out?
Dormir
yeah, I keep oracle software on it's own filesystem, crs on it's own, and log files on another
MxxCon
crs?
TheBonsai
i'm not that picky, but in 11g2 times, grid infra deserves itt, maybe
Dormir
so normal system activities don't fill up the oracle dirctories and I don't have to worry about full filesystems as much
TheBonsai
MxxCon: you want to build a RAC - google about crs before you start, please! :)
Dormir
crs = Cluster Ready Services = software you need for RAC
MxxCon
TheBonsai i'm not going to be the one actually setting it all up. i'll just install and config OS and hand it off to our dba
         

TheBonsai
MxxCon: oh. then it's easy: ask the dba for his specific requirements ;)
Dormir: do you know how well instance caging works (altering CPU_COUNT)?
Dormir
yeah, and if he doesn't know what CRS is, get a new dba :)
TheBonsai
heh
Dormir
no, haven't used it
well, I take that back. I think I used it years ago, but I can't remember why and haven't on any recent system
TheBonsai
i plan a general purpose cluster with a mix of single/racOne/RAC databases. i need at least *something* to control the monster :)
MxxCon
TheBonsai right now dba is n/a and i'd want to get this thing done before he's back
TheBonsai
MxxCon: with the effect that he says "unusable" and you can do it again (says Murphy..) ;)
Dormir
best way is to get the layout from the dba. you can allocate the VIPs and interconnect
TheBonsai
MxxCon: well, i'd be happy with one extra FS, Dormir wants 3. you'll get different answers, whoever you ask
Dormir
you can grab the install doc and make sure all of the patches are installed
but different dbas do things a little different
avaia
just make one huge /
make sure you use Gentoo, too
Dormir
like a shared oracle home for rac vs a local home on each node
TheBonsai
avaia: and don't forget to use the latest gaming machine
Dormir
as a side note, I like GenToo. :)
avaia
I do all my DBA work via wiimote
MxxCon
which is why i'm asking if oracle has something like that documented :D
TheBonsai
Dormir: i forgive you :)
Dormir
there are some white papers and the install guides
TheBonsai
MxxCon: i don't think so, because it doesn't matter for oracle itself. they document stuff like disk I/O for redo logs and so, but not if you start your software from USB stick, NFS or HD
MxxCon
what about floppy?
TheBonsai
if you have enough drives and a good logical volume manager, maybe :)
Dormir: a separate log fs isn't bad. a few months ago the warehouse did something like hexdumping sga contents into a trc
Dormir: 80G sga in a classic value/ascii format... O.R.A.C.L.E.!!!
MxxCon
(kinda offtopic but yesterday i was at mysql usergroup hosted at oracle hq listening to postgre speaker)
TheBonsai
heh
anyways, i'm going to sleep
Dormir
MxxCon: the power of open source :)
TheBonsai
bye all
Dormir
night Bonsai
TheBonsai: that's exactly why I like a seperate filesystem :)
TheBonsai
(Action) nods
MxxCon
every night 1000s of people die in their sleep. don't become a statistic, stay up.
xzilla
hmm, i was at that meetup too
Junior
yello ;)
markl_
Junior: kinda quiet this time of night
Junior
here it's 7 am :))
KeiKurono
morning
mgolisch
running orion on raw devices that are used by asm should not cause any harm right?
als long as i dont do writes
nunovsky
lol
mgolisch
whats so lol?
ill shutdown the database ofcourse before that
or will this destroy anything
any used orion?
i cant seem to get more than 7000IOPS out of my san, wheras the same test gives me like 19000 on my old san
odd
the datatransferrate with large IOs is more than double on the new san though
i wonder what i might have configured wrong
RJarett
mgolisch: more details
san type, spindles, how many SSD or PAM flash, how much cache, whats the interface type
mgolisch
the new san is a cx4-480
RJarett
eh
any ESD in it?
mgolisch
nope
RJarett
more details
mgolisch
2x 8gbit fc one to each sp
i think its 59 disks in on those multipath devices i ran the test on
15k fc
RJarett
and only 7k?
hmmm
do you have a very hot datafile or table?
mgolisch
no there is nothing installed on it yet
RJarett
i was able to push 7k+ through a netapp 3170 which is similar specs to 16 disks and a PAM card
how are you testing?
mgolisch
orion
RJarett
so not a real workload test?
mgolisch
might be something with the configuration of the luns and or the multipath config on the server
just wonder why the same test gives me like 4times the IOPS on my old cx700
RJarett
same cache size?
plus what filesystem on the LUNs?
linux i presume?
mgolisch
none
its to be used for asm
RJarett
whats your queue depth set to? which io scheduler?
mgolisch
where do i see that?
:)
RJarett
what os?
mgolisch
its redhat as 5
RJarett
what sd* device is this?
or other device name
mgolisch
its two multipath devices
RJarett
plus what HBA model/manufacturer
mgolisch
mpath*
RJarett
ok tell me both
mgolisch
qlogic 8gbit fc
/dev/mapper/mpath2/3 i wonder how i see what the actualy devices this is built from are
RJarett
cat /sys/block/sd*/queue/scheduler
mgolisch
noop anticipatory deadline [cfq]
for all
RJarett
echo deadline into each of them
whats you options qla2xxx in modprobe.conf?
mgolisch
there is none
RJarett
lsmod | grep qla
mgolisch
qla2xxx
RJarett
is this a production box right now?
grep -i "queue depth" /proc/scsi/qla2xxx/[0-9]*
if this isnt a production box: options qla2xxx ql2xmaxqdepth=256 qlport_down_retry=1 ql2xfailover=0 ql2xretrycount=5
but you may want to later tune the retry counts
grep "alias scsi_ho" /etc/modprobe.conf too
if you have an 8gb card you may even be picking up the wrong driver
may want to force which one its using
mgolisch
alias scsi_hostadapter mptbase
alias scsi_hostadapter1 mptspi
alias scsi_hostadapter2 megaraid_sas
alias scsi_hostadapter3 sata_svw
alias scsi_hostadapter4 qla2xxx
RJarett
what chip is in your HBA?
mgolisch
21:00.1 Fibre Channel: QLogic Corp. ISP2532-based 8Gb Fibre Channel to PCI Express HBA (rev 02)
RJarett
if its one of the 600 or 6000 series you can tell it to use a different family or setting within the qla2xxx driver
hmmm
mgolisch
also what i wondered about, shouldnt there be some config file in which the dm-mpio config is?
i see there is mpath* devices but the multipath.conf lists nothing
does that mean those devices will vanish if i reboot?
RJarett
honestly i dont think youre at that point
as far as perf tuning goes
man devmap_name
mgolisch
some dell guy configured it, iam just trying to understand what he setup and or why it performs that bad
:)
RJarett
ick dell
ya paid 40-60% more than netapp on an EMC plus it runs off a dell :P
from a high end to midrange san too ;)
mgolisch
i wasnt here when all this was installed/setup so iam kinda clueless
in the current production box we used powerpath , i know my way around that but i have no clue about dm-mpio
it just seemed odd that they planed to run asm over those two multipath devices, id have asumed that asm probably does a much better job managing the storage than the emc does, i think those mpath devices are built on two metaluns exported on the emc box
RJarett
asm doesnt do better for mpio
not until direct nfs did oracle handle it well
which doesnt apply to you
cmug
can i speed up a running import somehow?
other than fixing redo file sizes to avoid logswitches
RJarett
not a running one
cmug
yea I can't affect the import settings themselves, but something that I could do on the instance side ?
RJarett
not really
cmug
:(
RJarett
plus most import time is usually building indexes
cmug
this has been running for over 3 hours now, i'm not keen to stop it to try without indexes
i need to do the same import to another db though
RJarett
how much data to how many disks?
and is it an imp or an impdp?
cmug
3.5g, using netapp over nfs
imp
exp file from 9i, imp'n to 10g
r2
RJarett
ahh
3 hours huh?
cmug
yeah. a single table for that time
RJarett
sounds more like a network issue
linux box?
« prev 1 2 next »