Disk Enclosures
1748089 Members
4951 Online
108758 Solutions
New Discussion юеВ

Re: VA7100 - Lun creation?

 
SOLVED
Go to solution
Dan Alexander
Frequent Advisor

VA7100 - Lun creation?

I have just installed a va7100 on an L class server, and am ready to configure the LUN's...The array is full (15 disks) x 36gb. I have to move about 200GB of data onto it for starters. Anyone know the "best" practice for creating LUN's? should I just create a LUN big enough to accomodate the data, or create a bunch of LUN's and create a VG that way? Any performance issues? Does it matter with AutoRaid?

Thanks! Points will be given!
8 REPLIES 8
Brian M Rawlings
Honored Contributor
Solution

Re: VA7100 - Lun creation?

Some ideas:
1> Hope you have two paths to the array. This will let you alternate which is primary path for each LUN (in LVM, not in the array), for quasi-load-balancing.
2> Assign primary paths through controller 1 for LUNs created in RG1 (raid group 1), and primary paths through controller 2 for LUNs created in RG2. LUNs can be seen and accessed through either controller, but by the book you get better performance if you match them up this way.
3> Do not make just one big LUN for a VG. For load balancing reasons noted above, make at least two. Four is probably better, since this array supports lots of LUNs. The reason for at least four (maybe more) is that LVM creates internal structures for I/O, queues and buffers, etc, based on how many "PVs" are assigned to the VG. If you only have a few, you only get a few I/O resources from the OS. You want more than a few for each VG. For busy VGs, go for more than four, do six or eight.
4> Consider LVM striping across the LUNs you use to create a VG. This is done in the lvcreate or lvextend command, stripe size would equal the number of LUNs in the VG. This does 'round robin' I/O through all I/O structures and queues set up for that VG. In other arrays, it spreads I/Os effectively across spindles, but the VA arrays already stripe across all disks, so the benefit is mostly seen by utilizing all I/O structures within your VG.
5> Decide up front if you want Mirrored storage (raid 0/1), or Autoraid mode (current data is kept raid 0/1, old data is migrated to raid-5 when space is needed). If you make LUNs RAID 0/1, you can switch to Autoraid later, but you CAN'T switch from Autoraid LUNs to mirrored LUNs. Whatever you decide, all LUNs will be of this type. For optimal performance (at the cost of space), go mirrored. For optimal storage (at the cost of performance, which is still pretty good for most cases), go with Autoraid LUNs.

There's more, but it's all I have time for. Hopefully others will chime in.

Best Regards, --bmr
We must indeed all hang together, or, most assuredly, we shall all hang separately. (Benjamin Franklin)
Brian M Rawlings
Honored Contributor

Re: VA7100 - Lun creation?

A couple more things occured to me: I suggest enabling 'hot spare' mode, I like the improved availability (HA), it seems like a useful tradeoff. I also suggest turning on 'prefetch', it seems to aid performance for most envirionments. You can turn it on and off while you are running, to see if it helps in your case.

Check out this thread, it has several helpful ideas for those looking to set up a VA array:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x30e1e822e739d711abdc0090277a778c,00.html
We must indeed all hang together, or, most assuredly, we shall all hang separately. (Benjamin Franklin)
Ricardo Rocha
Valued Contributor

Re: VA7100 - Lun creation?

Hi

A small big correction to what Brian said. If this is a VA7100 there is only a RG. So you can only create Lun's in RG1. To have best performance, you must put the primary paths going through controller 1, otherwise you'll have great impact. Lun's bigger than 200GB are not that advisable, so you could consider to create 2 lun's with 100GB and then create a vg with these 2 disks. Before doing all this, update your firmware to HP18 (let HP do this), as this will protect more your data and increases performance.

Good luck,

Ricardo
"there is this old man who spent so much of his life sleeping that he is able to keep awake for the rest of his years"
Dan Alexander
Frequent Advisor

Re: VA7100 - Lun creation?

Thanks guys.

Brian, it is a VA7100, so there is only one RG...
But thank you both very much for the useful info...I appreciate the quick responses!

Dan
Brian M Rawlings
Honored Contributor

Re: VA7100 - Lun creation?

Right, my bad, should have read closer... thanks, Ricardo. I would still recommend more, smaller LUNs, however, for the LVM multiple queues and I/O structures that I outlined earlier. I don't know how much it helps, since all your I/O threads are aimed at one controller, but I'd do at least four LUNs per VG.

Regards, enjoy the easy-use array! --bmr
We must indeed all hang together, or, most assuredly, we shall all hang separately. (Benjamin Franklin)
Vincent Fleming
Honored Contributor

Re: VA7100 - Lun creation?

I'm told that the array performs better when you keep the LUNs under 2TB each...

Since 2TB is pretty big, that should not be a problem...

-Vince
No matter where you go, there you are.
Roger_22
Trusted Contributor

Re: VA7100 - Lun creation?

HP18 firmware, Prefetch enabled, Normal Resiliency mode, Queue Full Threshold set to 40. These are the best performance settings for a 15 disk 7100.

If you can afford RAID 1+0, then use that, its no risk best performance. The AutoRAID setting might work ok for you but you system is likely to be out of balance. A L-class server can easily over drive a 15-disk array. If the array is too busy, then the AutoRAID policies will suspend, and you???ll end up with a large RAID 5DP IO component ??? this will not be a performance winner.

If you choose the AutoRAID setting, then select the single active hot spare option. If you choose RAID 1+0 only and you???re interested in performance during a failed disk/rebuild scenario, then either 1) limit the total LUN capacity to one disk (36GB) less than the total available capacity, or 2) write a cron script to enable rebuild only during off peak performance times, like the middle of the night. These are only to minimize the performance impact of a failed disk, if that???s not a big worry, then don???t bother with these instructions.

Make a 10MB LUN 0 ??? this makes management in a SAN, if you ever upgrade to one, easier.

The array performance is not affected by LUN count or size. However, managing the performance is easier the fewer LUN you create. Create LUNs that match your application ??? in a logical organization.

Then, using scsictl, manage the HPUX queue depth. This will be very important. A L-Class server can be quite powerful; it may overwhelm a 15 disk array. Your task is to distribute the 40-slot queue to your LUNs. If it???s all a single application, then distribute the 40 queue slots evenly (ignore the management LUN 0). If you have a ???production??? application, and a lower priority application (like test), give more queue slots to the higher priority application.

Think of the ratios as a prioritization scheme. It???s ok to go over 40 by a little. If you???re interested I???ll explain more???.. good luck.

Roger_22
Trusted Contributor

Re: VA7100 - Lun creation?

HP18 firmware, Prefetch enabled, Normal Resiliency mode, Queue Full Threshold set to 40. These are the best performance settings for a 15 disk 7100.

If you can afford RAID 1+0, then use that, its no risk best performance. The AutoRAID setting might work ok for you but you system is likely to be out of balance. A L-class server can easily over drive a 15-disk array. If the array is too busy, then the AutoRAID policies will suspend, and you???ll end up with a large RAID 5DP IO component ??? this will not be a performance winner.

If you choose the AutoRAID setting, then select the single active hot spare option. If you choose RAID 1+0 only and you???re interested in performance during a failed disk/rebuild scenario, then either 1) limit the total LUN capacity to one disk (36GB) less than the total available capacity, or 2) write a cron script to enable rebuild only during off peak performance times, like the middle of the night. These are only to minimize the performance impact of a failed disk, if that???s not a big worry, then don???t bother with these instructions.

Make a 10MB LUN 0 ??? this makes management in a SAN, if you ever upgrade to one, easier.

The array performance is not affected by LUN count or size. However, managing the performance is easier the fewer LUN you create. Create LUNs that match your application ??? in a logical organization.

Then, using scsictl, manage the HPUX queue depth. This will be very important. A L-Class server can be quite powerful; it may overwhelm a 15 disk array. Your task is to distribute the 40-slot queue to your LUNs. If it???s all a single application, then distribute the 40 queue slots evenly (ignore the management LUN 0). If you have a ???production??? application, and a lower priority application (like test), give more queue slots to the higher priority application.

Think of the ratios as a prioritization scheme. It???s ok to go over 40 by a little. If you???re interested I???ll explain more???.. good luck.