XP512 questions

[Insu Kim]

Hi again,
One of your questions is,
In RAID 0/1 -> Where do the M2 (mirror) disks
sit? Is it predefined (like in
EMC). For example, is
Diskbay on L1 mirrored to a
diskbay on R2 ?

> Disks are sold in groups of four drives called array groups and RAID 1 copy is within the same array group.
For example, the first array group is installed in front bottom, up, over to back and down, then up like the letter N in the bottom of the frame of R1.
This is too difficult to explain everything so I recommend that you refer to the user's manual.

Hope this helps,

[Dave Wherry]

RajMan,
I'll take a swipe at answering some of these. My experience is just on the XP256. I assume it all applies to the 512 also.

Raid 0/1 or Raid 5. Disks are installed in an XP 4 drives at a time. This is called an Array or Marketing group. Over the years we've been told that Raid 5 will give you better read performance, but worse write performance. So, depending on your application, choose your raid level. If it's a read intensive application like Decision Support where people are doing a lot of lookups, use Raid 5. If it's a data acquisition system, stock transactions or gathering statistics, use Raid 0/1. A couple of years ago I got a white paper from Hitachi showing that their Raid 5 performs as well or better than Raid 1. It makes sense. All of your writes go to that massive cache. The disk latency is greatly reduced if not eliminated. Raid 5 performs very well.
So the next part of the Raid question is usable space. Remember you use 4 disks together. In Raid 0/1 you wil have 2 primary and 2 mirror disks. The best you could get is half the space available. With Raid 5 you have 3 data disks and 1 parity. More usable disk space, thus lowering your costs.
That array group that is installed is put in the same cabinet. Beyond that, they are put on the same ACP (bus). They can not be split up. So to answer your L1/L2 M1/M2 question, they are on the same ACP in the same cabinet. You can not split these up.
One other thing, you can not mix Raid levels on an ACP. You can have Raid 0/1 on one of the ACP's and Raid 5 on the other ACP. That was my original configuration. I have since converted everything to Raid 5.

You Open emulations is basically partitioning the drives. If you put an array group of 73GB drives in and format them as Raid 5, Open-8, you will get 29 logical devices (LDEVs), each ~7GB. 4 times 73 is 292GB. 29 times 7GB is ~203GB of usable space. When these drives are formatted, you lose a good deal of space in overhead. Just on a cost basis look for the emulation that gives you the most usable space.
As someone else said LUSE is similar to a metavolume. In Unix I prefer to use just the standard emulations and do host based stripping. For NT, we are using LUSE because those hosts prefer to see a larger device, rather than many small ones. A LUSE device can be up to 32 LDEVs and must be consecutive LDEVS. So they may be on the same array group, span array groups, or be on several array groups. It depends how the LDEVs are numbered/dispersed when the drives are formatted.
A CVS device is a custon volume. Your host needs a command device to talk to the XP. A command device is an LDEV that is only used as a command device. A utility called Volume Size Configuration allows you to slice off a smaller chunk of an LDEV to use as you please. Why waste a 7GB LDEV as a command device when you can make a Custom Volume (CV), say 64MB to use as the command device, leaving the remainder available for other use. Or, you only need a 1GB LDEV for an NT box. Create a smaller CV instead of wasting a whole 7GB LDEV.

...choice or flexibility in picking disks for an ARRAY GROUP?...
Short answer, no. Remember, 4 drives at a time in an array group.

I won't get repitive on the other questions. Others have done a fine jobs answering those.

[Roger Baptiste]

Thanks for the response. Some of my doubts seems to have cleared. Here is a summary of my intrepretation. Pls clarify , if i am wrong.

**
RAID -> I guess, i will use RAID 5, since the design of XP seems more tuned for RAID 5.
I don't see any point in going for RAID 1, when the disks and disk adaptors are shared by M1 and M2 and that too being housed in the same rack. (unlike in EMC ).

Array group -> pretty much predefined and there is nothing much which can be tinkered. Four disks all from the same rack
make an array group. Nothing more, nothing less?
Number of arraygroups = numberofdisks/4
(approx) ??

Emulation mode -> From the responses, i understand Emulation basically defines the
size of a LUN (which is got from the carving of array groups). I have to decide between
OPEN-9 and OPEN-E (ah! E in hex is 14 and that explains the 14 gb lun ;-) ). The size of lUN in turn defines the number of LUNS!.
Use only one Emulation mode for consistency and ease of maintenance!?

VG configuration -> I plan to pick LUN
from each ACP to create a VG. That way, i spread that VG load across all ACP's , disks, and channels.

VG-LV striping -> Not decided on it. 2 Mb
distributed extents is what i use on EMC (but that is with MEtas). Need to consider lv striping of 64 MB.

and ofcourse EMC <->XP translation is :-)
Disk director <-> ACP
FA director <-> CHIP
META <-> LUSE
VCM <-> CVS

Right?
***

Few more questions ;-)

What is a Control unit?? Is it ACP? or some sort of logical entity? How do we define it?
(or isit predefined? )

Any info on Sharedmemory? (do we need to tune or configure it?)

Is there something corresponding to EMC's "Rule of 17" in XP?? (Rule of 17 is the way the EMC's FrontEnd(fibre/scsi) directors are protected. Each FA is numbered and has a failover FA .

Also, in EMC i can see which FA (frontend director) a LUN is going through by using
symdev command. It helps in tracing the path of the LUN right from the system to the frontend. I don't see xpinfo giving that info. Is it the "PORT " info in the xpinfo output?


thanks again
-raj

[Bernd Reize]

Hi RajMan,

Array group -> exactly right

Emulation -> right, they differ only in their size. You have to make the choice between number of LUNs and size of one LUN. Using the same Emulation for all disks in one box should be the easiest way -- we didn't follow that and mixed OPEN-8 and -E's and got some maintenance problems out of it ;-)

VG config: exactly right. It is the optimal configuration to use a "round-robin" like way

your EMC<->XP translation is right, too


Control Unit: Just a term for internal adressing of LUNs (OPENs). One LUN is adressed by ControlUnit:LUN-Number. I think you could freely choice how to define CUs, but I would recommend using one CU per ACP. This way, the disks have all the same number per ACP, and one VG would i.e. use disks 0:00, 1:00, 2:00 and 3:00, the next VG would get the *:01 disks and so on.

SharedMemory: not configurable afaik. Tuneable is the Cache Layout: how much may be used for WriteCaching (in %). It is called "Duplex Write Cache Limit" I think. Depends of course on your workload, but may be worth tuning.

The XP Port Naming if different from EMC's: All Ports are divided into two halfs, the left and the right one. They are called CL1 and CL2 (Cluster) Within each Cluster there are the CHIP's, named A,B,C.. and so on. For ease of use/maintenance you should use them in pairs: i.e. CL1-A and CL2-A for HostA
You are right, you can see the used port in xpinfo output as 'PORT'

Bernd

[Duncan Edmonstone]

RajMan,

Remember that you can have a maximum of 256 LUNs per port, so if you have many systems redundantly acessing many disks and maybe many business copies too, you should use a larger emulation mode, or you may find yourself running out of ports.

HTH

Duncan

[Roger Baptiste]

Bernd, Thanks for your response. I guess, i have some broad idea now on the issues which needs to be addressed for configuring.

Will post back later with any further tuning questions, as i get started into it !

Thanks again everybody.
-raj

[Steve Bonds]

I found a great HP publication which describes in detail the OPEN-(whatever) specification.

http://www.hp.com/cposupport/manual_set/lpg28610.pdf

Found via Google search on "hitachi open-9 cylinders heads sectors". ;-)

To summarize (in case the link goes away):

------
Type sectors cylinders size (MB)
------ ------------ -------- ----------
OPEN-3 4806720 3338 2347
OPEN-9 14423040 10016 7042
OPEN-K 3661920 2543 1788
OPEN-8 14351040 9966 7007
OPEN-E 28452960 19759 13888
OPEN-3*n 4806720*n 3338*n 2347*n
OPEN-9*n 14423040*n 10016*n 7042*n
OPENK*n 3661920*n 2543*n 1788*n
OPEN-8*n 14351040*n 9966*n 7007*n
OPENE*n 28452960*n 19759*n 13888*n
------

ALL of these have 512 bytes per sector, 15 heads, and 96 sectors per track.

(I hope the crude ASCII formatting is readable after I post it...)

-- Steve

[Steve Bonds]

Well, the crude ASCII formatting is mangled. Try cutting and pasting into a text viewer that uses a fixed-width font.

-- Steve

[Vincent Fleming]

WOW - We've got a LOT of XPerts out there! It does my heart good to see everyone so helpful!

One thing to point out... about the shared memory...

Shared memory does take a little configuring... but this should be done by an HP technical consultant (TC - now called SA). The shared memory keeps all the housekeeping info, and needs to be bigger for bigger arrays (basically). You will need to increase SM if you use Continuous Access, Business Copy, want more than 1024 LUNs, and it's recommended for more than 16GB of cache.

Good luck!

[Dave Wherry]

Just an echo on Vincent's comment about getting HP technical help when configuring shared memory. I have an XP256 and we put in the maximum shared memory and cache. Maximum can't be a bad thing can it? Yes it can.
We also had 73GB drives and were doing Business Copy. To do all the Business Copy pairs we needed, we actually had to remove 2GB of cache.
There are several hidden gotcha's when configuring an XP so do work with HP. Also be sure to ask questions about upgrades and how that could affect the configuration in the future.

Dave