EVA 3000

[Daniel Pilloff]

I have recently installed an EVA 3000, four controllers, four drive cages, and 35 146GB drives. I was hoping that someone might be able to assist me with a few questions (I have tried to find these myself but haven't been able to find them in the documentation).

1. It appears and from what I have read that running the EVA 3000 in a single HBA single path environment requires that I zone off each individual connection of the EVA. Is this correct? The pictures I have seen of running the EVA have it connected to two separate fabrics. Can I run it in a single fabric by zoning off all the connections and only presenting a single connection on the EVA to each server (I then just balance the four connections among the servers).

2. I have VCS 3.0.10 (I think, I may be off a decimal point) and after going through the merry go round of different people I was able to acquire the Command View software on my SMA-II to manage it. I have initialized the storage and even presented a LUN. What I have run into is that when I originally tried to configure all 35 of my disk drives under one disk group the amount of usable space is only around 2 TB (VRAID5). However, when I configure two default disk groups 17 and 18 drives apiece, the amount of usable space is 1.7 and 1.8 GB (VRAID5). Is there a physical disk limit for the amount of disks in a disk group? I was under the impression that strictly for performance one disk group is the optimal configuration. The problem that I am running into is that I think I ought to see more usable space then 2 TB's when I configure the group this way. Any help would be greatly appreciated. Basically can I configure all 35 disk drives under one group and assuming I have the group set for single drive failure, what is the usable space I ought to expect?

Any help would be greatly appreciated.

[Stephen Kebbell · Accepted solution]

Hi Daniel,

First off, do you mean 4 controllers, or 4 FC-Ports over 2 controllers on the EVA3000? I would guess the latter.
1. You are correct for single-HBA implementation. Zone a single EVA port with each server. Don't forget a Management zone with the Appliance and one port of each EVA controller. See this doc from HP for Single Path implenetation:
ftp://ftp.compaq.com/pub/products/storageworks/whitepapers/5982-1850EN.pdf

2. Having an odd number of disks in an EVA is not recommended, but it will work. The largest LUN size you can create on an EVA is 2TB, no matter how big your disk group or whatever vraid level. All disks in the EVA can be in a single disk group - there's no limit, apart from the limit of how many disks the EVA itself can support (in your case, 56). For performance it's recommended to have just 1 disk group.
Setting the group to single drive redundancy means the EVA will reserve 2x the largest physical disk size for spare space = 288GB. For vraid1 you'll lose another 146GB for having an odd number of disks. So you'll have about 2.2TB of space if you use only vraid1 (50% raid overhead), and I'd guess about 3.8TB for only vraid5 (20% raid overhead). Also, 146GB disks have less than 146GB usable space. Mixing Vraid1 and Vraid5 in your disk group will give you usable space between those 2 limits.
Have a look here for EVA best practices:
ftp://ftp.compaq.com/pub/products/storageworks/whitepapers/EVA_Best_Practices_White_Paper.pdf

Hope this helps.

Regards,
Stephen

[Daniel Pilloff]

Thanks you very much for replying, this helps me a lot. I had read both documents you pointed to (I will go back and read them again) but I just want to make sure I understand what you stated about the disk group. When I configure a single disk group with all 35 of my disks the EMA reads as follows:

Capacity
Total: 4506.06 GB
Available:
Vraid0 2047.99 GB
Vraid1 2047.99 GB
Vraid5 2047.99 GB

From what you have stated I ought to read this as I can create a single LUN with the different various RAID levels with the maximum size of 2047.99. This 2047.99 is not the total amount of disk space I have available to create LUNS. That number is based on which raid level I use. VRAID1 is 50% or as you stated about 2.2 TB and VRAID5 accounting for its overhead will be around 3.8 TB (20% overhead). Am I correct in the assumption that how much disk space that is available for LUN creation is something I need to calculate as it does not appear on the appliance?
One more quick question, I currently have been managing 3 EMA 12000's for two years where the actual LUN number (D10 etc..) had a lot of significance, with the EVA, does the LUN number assigned to the VRAID have any significance?

Thanks again for replying.

[Sheldon Smith]

You can create a individual VDisk of up to 2000 GB. (And, since you are running VCS 3010, you should not create a VDisk larger than 1900 GB. This is a limitation in the VCS firmware and should be updated fairly soon.) When enough capacity is used, the Available Vraid0, Vraid1 and Vraid5 numbers will start to decrease appropriately.
When you create a VDisk, you tell it how much storage you want to appear at the host. It does the calculations and uses as much physical capacity as it needs to present the desired amount.
The LUN number simply sets the presentation order of the VDisks to the specific host.

[Uwe Zessin]

The HSG controllers have a global LUN address space. The 'name' of a unit also defines the LUN address. You can only 'shift' that range through CONNECTION offsets.

A virtual disk on the EVA is more like a container on the HSG. The EVA has more flexibility to assign a virtual disk to different LUN addresses of different hosts.

[Mike Naime]

With the EVA and AIX systems, I leave the OS UNIT ID field blank. It is assigned a unit/LUN number when you present the drive to a host. This is equivalent to the "D" number on the HSG's. When you select the PRESENT VDISK button, there will be a SELECT OS ID on the next page that allows you to choose the ID number instead of the Application assigning it.

With the EVA, you can have multiple hosts all with the same "LUN ID" being presented to the OS. You do not have to do and fancy connection offsets that need to be re-created if you have to delete the connection because the host hardware/HBA changed.

[Daniel Pilloff]

Thank you very much for your responses, they have been extremely helpful and having worked with three EMA's over the last two years I can really appreciate some of the changes with the EVA's that make storage administration considerably easier. One configuration I saw new with the EVA was that you can actually expand the raidsets. I tried to do this but the LUN presented to a Windows 2000 host remained fixed. Traditionally I have used multiple physical disks and Windows dymanic spanned volumes to increase logical drives. Is there any method to actually resize partitions to Windows 2000 hosts or is this an OS restriction?

[Uwe Zessin]

Daniel,
are you sure that the virtual disk has not been expanded? Note that the partition has a size recorded that, of course, cannot be changed with CommandView EVA. If you rescan the disks or reboot the server, doesn't Windows show some unpartitioned free space at the end of the disk?

I have not tried it myself, yet, but according to the documentation you need to alter the partition table with Microsofts DISKPART utility or use HP's "OpenView Storage Volume Growth" product.
http://h18000.www1.hp.com/products/storage/software/svg/index.html

The first method is described in the 'Installation and Configuration Guide' for Windows (AA-RUGZB-TE) on page 31.

If I recall correctly, you can expand a Raid-5 set on the HSG one time with another Raid-5 set by joining both together with a 'concatset'.


Mike,
I agree that host management has become much easier because you map a virtual disk to a LUN address for an entire host and can add/remove port wWNs without impact.

'fancy connection offsets' (have you really fallen in love with them? ;-) had multiple uses on the HSG:

- map another 0..7 LUN range from the HSG's global unit address space into the limited address range of a SCSI-2 driver. That allows different servers to access different units. E.g. connection offsets for server A are 0, so he can access units D0..7 and connection offsets for server B are 10, so he can access units D10..17 and need not share the range with A.

Some people have also used this to avoid LUN masking (A cannot address anything beyound D7 and B is limited to D10..17), but I hope they have fixed it before turning on large LUN adressing...

- allow a single fibre channel adapter with a SCSI-2 driver to access more than 8 LUNs through different controller ports by giving each connection a different connection offset. E.g. connection offset to Port_1 is 0, so FCA sees D0..D7 and connection offset to Port_2 is 10, so the same FCA sees D10..D17 here. I am talking about MULTIBUS_FAILOVER, here.

[Mike Naime]

Uwe:

I have a co-worker that would ALWAYS forget to check the AIX connections for Unit offsets. He would set something up, and I usually had to go "FIX" the connection offsets to make it work.

[Uwe Zessin]

Ah, that is frustrating, yes. I don't have any experience with AIX, so I wonder: why do you use offsets? Is AIX not capable of using 'large LUN' addressing?