HP Raid for OpenVMS

[Tom Monaco]

Is this still an actively developed product and does it support ODS5 disks? It looked like it had been ported to Itanium (though I couldn't find any pricing for that), but the only documentation I found online mentioned that it created ODS2 disks.

We have a standalone ES40 with a bunch of SCSI disks that we would like to assemble into larger units and this seemed like a straightforward way to accomplish that without changing hardware or giving up controller redundancy.

Are there any gotchas I need to look out for?

[Volker Halle]

Tom,

the SPD mentions both ODS-2 and ODS-5. As this product works at the device level, it really does not have to care about whether the file-system structure is ODS-2 or ODS-5.

SPD for HP RAID Software for OpenVMS V3.0A

http://h18000.www1.hp.com/info/SP4649/SP4649PF.PDF

Volker.

[Tom Monaco]

Thanks, I missed that.

I guess I need to figure out what will be required to use it on an Itanium server. I found a part number, BA383AC, but no other information.

[Robert Gezelter]

Tom,

I concur. RAID is far below the level of the actual file structure. It applies however one cares to make use of the disk (e.g., ODS-2, ODS-5, whatever).

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

[Hoff]

How big are the disks? Which disks are currently installed, and how much of the terabyte maximum disk limit do you need for your (actual or synthetic) disks here?

You mention Integrity, and you also mention an AlphaServer ES40. Depending on which of these (and which model Integrity), there are a variety of options available.

500 GB and 750 GB disks are available from HP and from other vendors, and can be configured on Alpha or Integrity with a little effort. (I configured a four-slot four-terabyte array for a client a week or so ago. It was about the size of a toaster. Cute, too.) One of the usual ways is an external shelf, though I'd tend to expect these drives would work inside whatever server has compatible slots (eg: rx2660, rx3600, etc).

RAID controllers are also readily available, either bus-based or outboard.

Host-based Volume Shadowing (HBVS; also known as mirroring or as software RAID-1) can be combined with controller-level RAID. HBVS can be a particular advantage as it allows replication of disks or of controller-based RAID or of virtual disks across hosts in a cluster.

Not that I'm looking to talk you out of host-based striping; out of acquiring and using HP RAID (FKA StorageWorks RAID, etc). Simply that you might want to investigate other options, as the disk product market is moving as fast as the rest of computing.

Stephen Hoffman
HoffmanLabs LLC

[Tom Monaco]

We have 24 disks (15K 18-36GB) in 2 split shelves (4354) connected to 4 dual channel U160 controllers (KZPEA). The disks are shadowed between 2 shelves and 2 controllers. Presently we are stripeing through the database (Ingres), but this requires some space mangagement effort we would like to avoid.

Our disaster-site system is similarly configured with older controllers and disks on a DS20.

Most of the alternatives using a smart controller or an MSA require a considerable amount of hardware reconfiguration that could affect the stability of the system, don't appear to offer as much redundancy, and I don't know that I'd want to maintain that configuration long term.

We are looking to begin a transition to Itanium sometime next year assuming all the required software is in place (CIFS, Ingres) and I'd like to address this issue before we start that.

We are likely to go with an rx2660 with either SCSI or SAS controllers for the disks. Since I've had controller failures before, I'm a little hesitant to use smart controllers with all the disks hanging off a single controller.

[Robert Gezelter]

Tom,

Thank you for the follow up information.

I would tend to be in the direction of using the controllers to do the actual RAID at the hardware level. Host-based shadowing (HBVS), referred to in RAID as mirroring, provides a way to use two different controllers.

With some care, and a small transition, it is possible to transition the disk data without interruption, as I described in my presentation "Migrating OpenVMS Storage Environments without Interruption or Disruption" at the HP Technology Symposium in Las Vegas earlier this year. The slides from that talk at available at: http://www.rlgsc.com/hptechnologyforum/2007/1512.html

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

[Tom Monaco]

Thank you for the reference.

Are you suggesting mirroring controller-based stripesets (raid0) with HBVS, or mirroring 0+1 stripesets and doubling the number of disks?

It doesn't appear that HP Raid for OpenVMS supports dynamic expansion of the arrays like the controller-based raids do, so that could be a limitation going forward. I guess I shouldn't expect this to change in the future.

Do you think using software raid (Raid1 on top of HBVS) would impose any significant overhead on the system, or are we well past those days?

[Guenther Froehlin]

An incoming QIO for the DPA device is split into individual I/Os by DPDRIVER using in-memory meta data. DPDRIVER makes a fairly simple calculation to decide how to split the user I/O and alloactes I/O Request Packets (IRP) accordingly. The IRPs are then queued to the individual disk driver which does the I/O work. DPDRIVER is not doing the physical I/O. Tests I did long time ago showed kernel mode overhead <3%.

Dynamic expansion of DPA devices is not planned. Due to the layout of the container file (meta data per container layer is allocated at its top end) this is nearly impossible to do.

/Guenther