Va7410 space allocation

[Diego Balgera]

Hi,

(newbie question in disk array environment).

I have a va7410 disk array with 40 disks and total size of 1335Gb.
What I would like to do is to create a single Lun to be used in a single volume group and allocate in this vg a single logical volume with a vxfs file system.

Due to the big size of this disk array, is this possible and/or a good idea? Do you think that there could be technical limitation in allocating such space or is a better idea to split it, considering also the performance point of view?

I'm asking this because other colleagues split the available space into different Lun/Vg/Lvol but I could not understand the reason.

Of course, it could depend also on the final usage, but just to have an idea of the feasibility ...

Thank you in advance.
Best regards.
Diego.

[Karthik S S · Accepted solution]

Hi Diego,

It depends ... If your application requires (such as a huge database) more space then there is no point in splitting the available capacity into multiple LUNs and VGs. Other problem of having a single huge LUN is that only one host can have exclusive access at a time there by you can not effectively share the storage capacity.

Assume that you have connected multiple hosts to your VA7410 using a FC switch and all the hosts wants to utilize the storage. Then you should be creating multiple LUNs and assign them to hosts (by default it will be available to all the hosts).

Even in a single host scenario you might want to create multiple LUNs and VGs depending on your applications and data needs. For example if you are running MC Serviceguard cluster with four packages (applications) you have to create 4 LUNs and 4 VGs.

The advantage on having multiple VGs is manageblity. If you want to perform some maintenance on one VG (by deactivating it) you can do so without affecting the access to the other VG.

Hope I didn't add more to your confusion ;-)

-Karthik S S

[Diego Balgera]

Thank you for your answer,

you are right.
The usage of the disk space is very simple, the disk is used only for file server purposes. There are no application requiring exclusive access, so there are no drawbacks in creating a single lun/vg/lvol under this point of view.
On the contrary, we want to maximize the space availability and creating a single file system will help.
But I wondered if there were other limitations in creating so big partitions and if this could be a drawback for performances (slow disk access).

Thank you again.
Diego.

[Karthik S S]

Hi Diego,

Since you will be using it for file server purpose I would suggest you to create 2 LUNs on the VA (associate the LUNs to different controllers so the IO is evenly distributed across the VA controllers) and create a single VG clubbing both of these LUNs.

-Karthik S S

[Peter Mattei]

You cannot create one big LUN on a VA7410.
The VA7410 has two redundancy groups. RG1 owns all odd disks (1, 3, ... 15 of all enclosures) and RG2 owns all even disks (2,4,...14 of all enclosures).
So you must create at least two LUNs to be able to use all space.

Advantages:
- less administrative work

Drawbacks of doing that:
- You only have 2 IOQueues, Unix likes to have multiple IOQueues for performance
- Backup / restore: If a large LUN goes corrupt by any reason it will take a long time to restore or do fsck. Small LUNs take less time

If you want to know more about how the VA works, have a look at this manual starting on page 13
http://h200005.www2.hp.com/bc/docs/support/SupportManual/lpg60187/lpg60187.pdf

Cheers
Peter

[Roger Buckthal_2]

With HPUX you also need to be aware to system queue depth. By default HPUX allows only 8 IOs to be active to a LUN at one time. When you have either large LUN or a very few LUNs on HPUX you need to match sure the queue depth is sufficient to keep the array busy.

Just think about it. If you have only one LUN for a 20 disk RG (easy to do on the VA), and each of the 8 HPUX IOs has gone to a different disk, you still have 12 disks that are idle. Not a very good utilization.

Although a simple concept, the complexities occur when you have multiple LUNs and multiple systems sharing the same array or LUN.

My basic rule is 2 or 3 IO queue slots per disk. More is better, but too many can also cause problems. When HPUX sends an IO to the array, it starts a timer. If the IO does not complete within the time-out value, it will cancel the IO and re-transmit. If this fails enough, other re-tries or failure recovery methods are activated. So, if you give the array more IOs than it can process in the time-out period, youâ ll create more work for the array.

You can see the HPUX queue depth per LUN with the scsictl command. Unfortunately, it is not persistent thru re-boots.

scsictl -a /dev/rdsk/c3t5d0 - display the current value
scsictl -m queue_depth 4 /dev/rdsk/c3t5d0 - set a new value to

[Diego Balgera]

Hi smart guys,
thank you again for your valuable suggestions.

To summarize: it is better do not create a single Lun for the above reasons. But what about the following proposal:
- creating several (i.e. 8) Luns,
- associate each of them to a physical volume
- create a single volume group with all the Pv
- create a single logical volume in this volume group

This could allows to use the Io queues properly and optimize the disk usage (my goal is to obtain a large file system possibly without affecting the performance).

Thank you again!
Best regards.
Diego.