New System and Storage Questions

[Charles McCary]

Group, hi - we're getting in 2 new rp7420s (one of which we're going to hard partition into two different machines) and a new EVA3000 disk array and I have a couple of questions for you to make sure I'm not missing anything...

1) Are there any root file system size recommendations with 11.1? I'm looking at my current configuration on 11.0 which for the most part has been ok and I'm figuring on just increasing the vg00 filesystems that have had space issues...anything else I need to know about?

What about file system size in general (outside vg00)?

2) With the EVA3000 are there any recommended LUN sizes? The reason I ask is that I currently have a couple of volume groups that will have 800 gb or so in them and some filesystems within those volume groups that will be 100gb or so, so I'm wondering should I have 50gb LUNs, 10gb LUNs, or does it matter as long as I'm conscious of load balancing? And since we're buying securepath, do I even need to be conscious of that...?

Thanks for your input...

[Steven E. Protter]

Gonna do an A. Clay immitation.

The answer depends.

The standard recommendations from Ignite for 11i installation will work reasonably well. I generally bump up /var a few GB because its a pain when that fills up.

The answer to question 1 depends on how you use the system. Here is a vg00 bdf for my systems. They run heavy oracle databases and have a lot of applications on them for security, patching and other issues. I'd go larger on /usr and /opt, they are pretty full now and thats becoming annoying.

Filesystem kbytes used avail %used Mounted on
/dev/vg00/lvol3 1048576 184175 810574 19% /
/dev/vg00/lvol1 512499 81167 380082 18% /stand
/dev/vg00/lvol8 6144000 1674831 4191969 29% /var
/dev/vg00/lvol9 2097152 671518 1336553 33% /utmp
/dev/vg00/lvol7 2162688 1970826 179884 92% /usr
/dev/vg00/lvol6 1048576 106211 884820 11% /tmp

8 GB of RAM

swapinfo -tam
Mb Mb Mb PCT START/ Mb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
dev 4096 0 4096 0% 0 - 1 /dev/vg00/lvol2
reserve - 1282 -1282
memory 6509 1021 5488 16%
total 10605 2303 8302 22% - 0 -


I believe 8 GB of swap would be optimal.

2) LUN size is totally dependent on what kind of data you are sitting on it. I recommend raid 1, raid 1/0 for oracle data and large databases.

SEP

[Geoff Wild]

As an example, this is how I setup my root:

Filesystem kbytes used avail %used Mounted on

/dev/vg00/lvol3 204800 140608 64192 69% /
/dev/vg00/lvol1 298928 72304 196728 27% /stand
/dev/vg00/lvol8 4710400 2514248 2181480 54% /var
/dev/vg00/lvol13 2097152 134779 1839728 7% /var/opt/perf/datafiles
/dev/vg01/lvol2 12582912 3688 12186130 0% /var/adm/crash
/dev/vg00/lvol7 2097152 1564784 528864 75% /usr
/dev/vg00/lvol14 524288 147958 354271 29% /usr/openv
/dev/vg00/lvol4 524288 72328 449928 14% /tmp
/dev/vg00/lvol6 2097152 1622424 471056 77% /opt
/dev/vg00/lvol12 262144 6275 239915 3% /opt/cctool
/dev/vg00/lvol10 24576 1830 21384 8% /ops
/dev/vg00/lvol5 524288 157760 363688 30% /home
/dev/vg00/lvol11 1048576 171447 822365 17% /data
/dev/vg00/lvol9 1572864 717339 802095 47% /app

Check out "Disk Space and Memory Requirements" under "Installing and Updating System Requirements " in "HP-UX 11i Version 1 Installation

and Update Guide":

http://docs.hp.com/hpux/onlinedocs/5990-7279/5990-7279.html

Rgds...Geoff

[Charles McCary]

Stephen,

thanks for the info...we'll have to use raid5 due to space requirements.

What's your reasoning behine raid 0 or 0/1 for large oracle databases?

thanks,
c

[Steven E. Protter · Accepted solution]

Oracle has specifically recommended that data files that have a lot of changes be on Raid 1 or Raid 10, because this provides better performance.

These recommendations are echoed for this kind of data in Chapter 7 of Charles Keenan's HP-UX CSE guide.

Because of the way data is stored on disk in the Raid 5 format, it is impacted by a lot of i/o all over the disk.

Raid 5 is fine, for write once read many situations, because the data doesn't get fragmented, making the subsequent reads and writes more complicated.

Keeping your filesystems defragmented with Online JFS can help.

That about summarized things. We go Raid 10 with Data and Redo logs.

SEP

[Mark Grossman]

Charles,
focusing more on question 2:

We have some disk limitations as well. We have an EVA 3000 running Oracle databases and Oracle Applications.
The actual database itself is probably around 125GB per instance. The apps add more to this and we also have a second web apps server.We use business copy for snapclones quite a bit to clone our instances for test and dev, so we divided our eva virtual disks/luns with cloning in mind as well.
Our layout is:
DB server:
Lun1 - 220GB Raid 5 contains database and oracle apps(infrequent updates). This is used for cloning an instance.
Lun2 - 25 GB Raid 1 contains redo, undo, temp databse logs. This is used for cloning an instance.
Lun3 - 230GB Raid 5 contains generic disk space for staging, tarring, etc This is not used for cloning instances.
WEB server:
Lun1 - 30 GB Raid 5 contains web apps for oracle. This is used for cloning instances.

We are very happy with the performance of this layout even though it is using Raid 5 and invokes the 'raid 5 write penalty'.
Of course we went from scsi Autoraid to Fiber EVA, so to us it is absoulutely flying. I didnt see much difference in performance when testing lun sizes.

I also tested the 3 options for Secure Path for load balancing. Again, our tests showed no difference between the 3 methods and we just let it use the Round Robin method.

Again, this is all just my experience so far with the system, so please dont all you gurus start telling me I'm nuts (just joking)

Mark

[Alzhy]

1) The 7420 has 4 Internal Disks (2 each side of the CoreIOs. They could be 36GB or 73GB pluggables. If your 2 nPars will be setup with no High Availability - then yo'd probably go with just using the 2 internal disks. If not then you're configuration would probably involve 2 DS2300's or 4 DS2100's so your boot disks are spread on 2 enclosures. Or you opt for SAN Boot disks. Which ever which way, I suggest a minimalist VG00 and use of the BOOT/OS disks exclusively for OS/System stuff. My partitioning is as follows (for all kinds of servers..):

stand 512
root 8192
opt 8192
var 8192 (or even bigger)
tmp 1024

what remains.. you can use for swap.
If you've the need, set up other swap on other pairs of disks or on your SAN disks if supported.

2) Use large LUNs and standardize. Your EVA system will like it better. We have 3 sizes.. 50, 100 and 200 GB. Up to you if you ant a Fileystem on each LUN or stripe a filesystem accross 4 or 8 LUNs.