Why use distributed stripes vs KB Stripes

[Ryan B]

Hello~

We have an admin that left our company and this person used distributed striping vs KB striping.

We understand how you set it up and how to tell if it is setup, but we need to know WHY you would use distributed stripes vs KB Stripes? The servers involved are N-Class and attached to an XP512.

[Duncan Edmonstone]

Generally there are 2 reasons to use ectent based striping:

1. You can mirror extent based striping, you can't mirror 'normal striping'. Not much of an issue on a disk array like an XP, but might come in handy if and when you migrate to another array.

2. With normal striping, if you were striped across 10 LUNs, and you used all the available extents on those LUNs, when you wanted to extend the LVs, you would need to add another 10 LUNs to the VG to let you do it!. With extent based striping you could add as few as 2 extra LUNs and extend the LV. Of course you might get different performance characteristics for that bit of the LV, but at least you have the flexibility to do it.

HTH

Duncan

[Ryan B]

Duncan,

Thanks for the info, and I do have a follow up question to clarify. Our current setup from VGxx is:

--- Physical volume groups ---
PVG Name PVG0
PV Name /dev/dsk/c14t13d3
PV Name /dev/dsk/c12t13d4
PV Name /dev/dsk/c13t13d4
PV Name /dev/dsk/c15t13d5
PV Name /dev/dsk/c14t13d5
PV Name /dev/dsk/c12t13d6
PV Name /dev/dsk/c13t13d6
PV Name /dev/dsk/c15t13d7
PV Name /dev/dsk/c14t13d7
PV Name /dev/dsk/c12t14d0
PV Name /dev/dsk/c13t14d0
PV Name /dev/dsk/c15t14d1
PV Name /dev/dsk/c14t14d1
PV Name /dev/dsk/c12t14d2
PV Name /dev/dsk/c13t14d2

Then the lvol looks like this:

Bad block NONE
Allocation PVG-strict/distributed
IO Timeout (Seconds) default

--- Distribution of logical volume ---
PV Name LE on PV PE on PV
/dev/dsk/c14t13d3 1667 1667
/dev/dsk/c12t13d4 1667 1667
/dev/dsk/c13t13d4 1667 1667
/dev/dsk/c15t13d5 1667 1667
/dev/dsk/c14t13d5 1667 1667
/dev/dsk/c12t13d6 1667 1667
/dev/dsk/c13t13d6 1667 1667
/dev/dsk/c15t13d7 1667 1667
/dev/dsk/c14t13d7 1667 1667
/dev/dsk/c12t14d0 1667 1667
/dev/dsk/c13t14d0 1666 1666
/dev/dsk/c15t14d1 1666 1666
/dev/dsk/c14t14d1 1666 1666
/dev/dsk/c12t14d2 1666 1666
/dev/dsk/c13t14d2 1666 1666


What I was told, is when we run out of space to extend this particular lvol, we would have to add 15 more luns to the VG/PVG to accomplish and extend. You mentioned, with the caution of a potential change in performance, that we could just add 2 more luns vs 15 more to do the extend. From what you see with our setup, is that still true??

Thanks for the help!

[TwoProc]

There are some papers out there (from HP) that recommend extent based striping for even large storage servers (like XP). It's goal is to stripe across many extents, but due to latency of small data gets from disk - it seeks to grab large amounts of data per disk seek. Theory here is increased throughput, application examples are for Oracle databases.

[Duncan Edmonstone · Accepted solution]

Ryan,

Yes, looks to me like just adding another 2 LUNs to the VG would allow you to extennd the LVs. I wouldn't of course reccomend it though - another 15 LUNs would be more appropriate.

HTH

Duncan

[Ryan B]

Thanks Duncan.

That helps clarify. One final thing for you or anyone else following this post.

Can anyone point us to any of these papers that discuss or recommend extent based stripes with large arrays that John was talking about???

Thanks!

[Duncan Edmonstone]

Ryan,

Assuming your ITRC ID is linked to an active support contract, you should be able to view this:

http://www4.itrc.hp.com/service/iv/docDisplay.do?docId=prodITRC/DE_SW_UX_swrec_EN_01_E/XP-DiskArrays.pdf

HTH

Duncan

[Ryan B]

Duncan~

It looks like this link if for the European site. Do you know how I can make this work for the Americas site OR would you be able to send it as an attachment to the forum or is it too big?

Thanks!

[Duncan Edmonstone]

Well here's how I get to it:

1. Log in to ITRC

2. Under " maintenance and support (hp products)", I select 'navigate knowledge trees'

3. Under HP-UX select "HP-UX Software Recovery Handbook"

4. "Chapter 26 - XP Disk Arrays"

I can't post protected material here - a forum moderator would just remove it.

HTH

Duncan

[Ryan B]

Duncan~

Thanks for all your help!!!!

We actually had a contractor in that setup a login the the European Site a long time ago and he got it for me.

Thanks Again!!

Ryan

[Duncan Edmonstone]

Ryan

Thankyou!

Your points just pushed me over 7500 - I'm just admiring my new headgear!

Duncan

[Ryan B]

All questions resolved.

Closing Thread

Ryan

[generic_1]

Use an extent size of 1024. Cavre your lun sizes small ahead of time on the XP so that that are layed out evenly across your controllres and allocated in order so you are striping eveningly across the back of your frame/disks/controllers/and fibers. If you plan your data layout well enough you really dont need powerpath on an XP which will save you some money, and CPU overhead. Small luns give you better read ahead, caching, and are less wasteful because of their small unit size.