logcal volumes (partitions?) on OpenVMS

[BoyeDav]

I know "partition" is a DOS/Windows/Linux term, but it's the best way to describe what I'd like to accomplish.

We have a P400 SmartArray controller card with (8) 73GB SAS drives. I plan to mirror two of the drives to use as my boot/system volume (using hardware mirroring). The remaining six drives would be configured as a single RAID 10 volume (using hardware RAID).

I'd like to subdivide the RAID 10 volume into 3-4 logical volumes, and I don't think there's any way to do that at the controller level. In order for OpenVMS to be able to use that RAID 10 volume as several logical volumes, do I need to use LDDRIVER, or is there a better approach?

[BoyeDav]

We're running OpenVMS 8.3 on Integrity rx3600.

[Jess Goodman]

Have you considered using rooted logical names? This is a VMS approach that allows you to make a top leve directory appear like a entire device. For example:

$ CREATE/DIRECTORY DISK$RAID1:[USERD]
$ CREATE/DIRECTORY DISK$RAID1:[DATAD]
$ DEFINE/TRAN=CONC USERD DISK$RAID1:[USERD.]
$ DEFINE/TRAN=CONC DATAD DISK$RAID1:[DATAD.]
$ CREATE/DIRECT USERD:[MYDIR]
$ COPY MYFILE.DAT USERD:[MYDIR]

The difference is that rooted directories share the entire volume's disk space and other resources while each partition has its own hard limits.

[BoyeDav]

Yes, but I appreciate you mentioning it anyway. What I like about that approach is that we lose the overhead of LDDRIVER.

[Hoff]

OpenVMS does not now support any form of drive-level partitioning (all use of the GPT aside), the classic approach toward that end involves a virtualizing storage controller.

Based on a very quick look at the Smart Array P400 technical documentation, it does appear that the controller can instantiate logical drives to the host; see the P400 support manual. This logical drive detail is rather murky in the manual, however.

http://docs.hp.com/en/J6369-90037/index.html
http://h71000.www7.hp.com/openvms/products/acuxe/index.html

It's certainly worth a try...

The other alternative involves purchasing more drives.

The LD I/O overhead is comparatively low (as it uses a back-door access path into the device driver), and is the typical solution here, if rooted logical names are not used.

Applications can and do typically see concealed rooted logical names as devices, and disk quotas can restrict storage.

Virtualized storage: http://64.223.189.234/node/408
RAID-5 and RAID-10:
http://64.223.189.234/node/410

[Uwe Zessin · Accepted solution]

> it does appear that the controller can instantiate logical drives to the host

Absolutely! I do have two P400 configured with two logical disks, each. On ProLiant DL380 G5 with VMware ESX server, but I am sure OpenVMS will accept it, too.

> This logical drive detail is rather murky in the manual, however.

I don't have time to look into a murky manual ;-)
but it is quite simple with ACU:
- choose a set of disk drives that form a 'disk drive array', sometimes simply called array
- create the first logical disk within the array by specifying size and RAID level
- continue until you are satisfied or the array's capacity is used up or the max. number of logical disks has been reached

[Jan van den Ende]

BoyeDav,

I do not know the specifics of the P400, but iur HSG80s used to, and the EVAs do, present chunks of their RAID sets as drives to VMS.
We just specify the desired size, and our SAN guys present that amount. And having set up DVE, whenever we need more, we (arrange budgets and all that and) ask the SAN guys to extend the presented volumes, and they do, and we
$ SET VOLUME/SIZE
et voila!

So, from the VMS point of view, it definitely works. And according to Uwe, the P400 can manage as well, so: go ahaead!

Proost.

Have one on me.

jpe

[Robert Gezelter]

BoyeDav,

I would take a careful check of the documentation on the controller. I have seen many times where it was assumed that the controller could do far less than it was in actuality capable of.

Using LDDRV is an excellent approach. I would not presume that the overhead is an undue burden. Do not presume, do the science (e.g., create the volumes as a test and run the resulting configuration against a "real" drive to compare the performance. When doing this, be sure to verify that the drive is setup to allow full windows and that the LD container files are contiguous (otherwise, the benchmark is measuring the ability of the XQP to process window turns, not the overhead of LDDRV).

I have also been using the rooted concealed logical name approach in conjunction with quotas for many years (very nearly 30). It is high effective, even in the face of use communities that do not cooperate with each other (read that as "hostile").

The quotas prevent one group from seizing the entire supply of free space. It is also possible to overallocate the quotas, but in that event I advise careful monitoring, it is easy to have a crisis.

- Bob Gezelter, http://www.rlgsc.com

[John Gillings]

Boye,

>The remaining six drives would be
>configured as a single RAID 10 volume
>(using hardware RAID).
>I'd like to subdivide the RAID 10 volume
>into 3-4 logical volumes

From a reliability and performance perspective, you'd be MUCH better off taking your 6 drives and forming 3 independent RAID 1 sets, then presenting them to OpenVMS as as 3 independent volumes. In terms of performance, you don't have the I/Os of 3 logical volumes tripping over each other (particularly in fighting each other for seeks), and for reliability, the failure of a single drive only affects a single volume. Moreover you can tolerate up to 3 drive failures (as long as each is from a different volume).

Even better, present all 6 drives to OpenVMS and use host based volume shadowing to form three shadow sets.

Apart from RAID1 (mirroring), all other RAID types were a great idea when disks were small and expensive, but now that they're large and cheap, RAID is a technology past its used by date.

Think about it, RAID10 is 0+1 right? so you use hardware to glue three drives together (RAID0), then mirror them (RAID1). Now you want another mechanism to break them back into three? Why not cut out the two middle men, save a whole lot of work, and eliminate several potential failures?

[Jur van der Burg]

>Using LDDRV is an excellent approach. I would
>not presume that the overhead is an undue
>burden. Do not presume, do the science (e.g.,
>create the volumes as a test and run the
>resulting configuration against a "real" drive
>to compare the performance. When doing this,
>be sure to verify that the drive is setup to
>allow full windows and that the LD container
>files are contiguous (otherwise, the benchmark
>is measuring the ability of the XQP to process
>window turns, not the overhead of LDDRV).

The overhead of LDdriver is minimal, it just has to pass along a request to the underlaying driver. For efficiency, LD containerfiles do not need to be contiguous, and the additional overhead if they are not is minimal as they are opened with cathedral windows. The XQP is never involved with LD once the containerfile is connected to a drive, the driver handles the mapping itself.

Fwiw,

Jur (LDdriver author)

[Robert Gezelter]

Jur,

Thank you for weighing in on my comment. And yes, I was referring to ensuring that "cathedral" windows are in use.

However, I still stand on my general comments about benchmarking: be careful what you benchmark, one may not be measuring what one thinks is being measured.

- Bob Gezelter, http://www.rlgsc.com

[Verne Britton]

OpenVMS has a layered product

HP RAID Software for OpenVMS (SPD 46.49)

that (I think) takes a disk and makes one or more really big container files on it, then presents those to VMS as logical disks ... WOW that sounds a lot like LD :-)

Never used it, just know it exists (especially useful for sites that must have everything fully vendor supported).

Of course being a layered product, there probably is a license cost to it ...


Verne

[Karl Rohwedder]

HP RAID SW creates software based RAID-5 or RAID-0 out of a bunch of disks. If you have also shadowing licensed, it works together with shadowing and creates RAID 0+1 arrays.
The arrays can then be cut into smaller pieces.

We are using it a lot.

regards Kalle

[BoyeDav]

Thanks everyone for the input.

My current inclination would be to let the P400 controller subdivide the array into multiple logical volumes. While I'm certain both LD and host-based shadowing are low-overhead, I think I'd prefer to just let the hardware handle it and have VMS just see the logical drives as separate devices. Otherwise, I'll probably just use rooted logicals.

[Robert Gezelter]

BoyeDav,

Also, remember to print out a copy of the configuration against the possibility of a hardware problem at some future date.

- Bob Gezelter, http://www.rlgsc.com

[Dean McGorrill]

I'd definitly mirror whatever drives you
finally decide to configure as John suggested. I'm regularly losing drives and saved by the mirrors! Dean

[GuentherF]

HP RAID SW can be used with a single drive (as seen by VMS) using RAID-0. You can create up to 64 logical drives (or partitions) which behave like a regular disk.

And for some redundancy the single drive can be shadowed using HBVS to create a RAID-0+1 array.

/Guenther