HP DL series true hardware raid with Linux

[Matthew Darcy_2]

guys

what is the smallest HP DL class server that currently has true hardware raid, I've got a ton of G3/4/5 360's that use the hardware raid cciss driver in the kernel fine, however the DL230 G6 that's just turned up boast hardware raid but requires specific kernel modules to see them - ala fake raid.

What's the current DL range raid status for true hardware raid ranging from low end to top end ?

[Matti_Kurkela · Accepted solution]

The "DL range" is not one-dimensional: a fully-outfitted DL180 G6 is easily more capable than an entry-level DL320 G6. Some models can be ordered as either "Entry/High Efficiency" versions or "Performance" versions - sometimes this makes the difference between "SmartArray B110i SATA RAID controller" and a _real_ SmartArray controller.

I tend to use HP QuickSpecs for finding answers to questions like this:
http://www.hp.com/go/quickspecs -> Worldwide -> HP ProLiant servers -> Proliant DL series (rack-optimized), then pick the model you want information about.

In the "Overview" chapter, look in the "At a Glance" sub-chapter and the "Storage Controller" bullet point. If it's SmartArray Pnnn or Ennn, it's a real SmartArray; if it's "SmartArray Bxxx", it's a "fake raid".

MK

[Steven E. Protter]

Shalom,

Most DL servers have the option of a P400 series RAID card. This is hardware raid. It has its own disk controller, memory and CPU.

It is not fraud or any other kind of chicanery.

It requires specific kernel modules for the OS to see the RAID because its hardware raid. the model HP uses to present the disk makes the raid structure invisible to the OS.

This is IMO a good approach that avoids confusion.

SEP

[Matthew Darcy_2]

both providing great feedback, however the technology that requires a specific kernel module to be able to see the disks on the controller in a raided configuration (other wise it just see's it as two disks) is fakeraid - that's not a derogatory term, just what it's known.

I'll track down the servers I need with a true hardware raid rather than fakeraid.

[Steven E. Protter]

You have every right to be wrong on this and post it to your hearts content.

The bottom line is the kernel modules are required because its true hardware raid. Deny it all you want. I don't make any money or gain benefit from disagreeing with you. Apparently though I'm denied ITRC recognition by you for disagreement.

Truly a violation of the spirit of ITRC.

Good day Sir,

SEP

[Matthew Darcy_2]

sorry, I'd not assigned points to you yet, and will do now that was not an intential snub, I didn't assign points to the first post until I'd had chance. tt's all valid input all the same.

the point I was badly putting across was that the cciss controllers of earlier models (the one I'm referencing as true raid) didn't need software to see the raid disk, they just saw the controller and delt with whatever the controller presented them, which is more in line with a hardware raid controller. The later definition on the HP kit that I was referncing as fakeraid is more in line with what is on a home users motherboard where as you rightly say the raid disk is presented at a software level on the hardware controller, eg: the actual raid presentation cannot be seen without a specific software module rather than a module for the disk controller as a device.

I am trying to avoid this as the the modules for these devices are not always compatible with updates from companies such as Redhat, and depending on a 3rd party module to be able to see the core operating system is not acceptable to me at this time hence looking for true hardware raid.

[Steven E. Protter]

Shalom again,

I see your point:

>>>
I am trying to avoid this as the the modules for these devices are not always compatible with updates from companies such as Redhat, and depending on a 3rd party module to be able to see the core operating system is not acceptable to me at this time hence looking for true hardware raid.
<<<

I've operated HP DL class servers fat and happy without the OS having any knowledge of the underlying RAID setup.

PSP for DL class servers also provides a glitchy interface on a web server to gather this information. There is no need to use the PSP kernel modules in most set ups, therefore you can continue to update the system from Red Hat.

Your concern is absolutely valid, I'm just not agreeing with your characterization of the RAID as fake. It's handled disk failure for me a number of times. I'm sure there are still dozens of operating DL class servers at NDS Jerusalem and elsewhere across the organization that I configured with hardware raid and are up and running, even after disk failure.

Isn't that what raid is for. I just can't make the leap to call the RAID fake, simply because you need a kernel module to see it from the OS. But I do agree with your concerns about using the Kernel modules.

To comply with SOX, it is necessary to update the systems, periodically, and that can break things.

SEP

[Matti_Kurkela]

> the cciss controllers of earlier models (the one I'm referencing as true raid) didn't need software to see the raid disk

To me, this is a bit problematic way to tell "true RAID" from "fake RAID": after all, without the "cciss" kernel module you don't see neither the controller nor the disks.

I think it would be possible to make a true RAID controller that would also have a standards-compliant (AHCI or whatever) interface to the disks. The standard interface would offer basic functionality, and the hardware-specific interface would be used to activate hardware accelerator features when desired.

This would make implementing drivers for the controller design much easier, as the OS installer could use the built-in standard interface, and after the OS is up and running, an extra driver could be easily installed to activate the acceleration features (although I expect the "handover" to be tricky).

It might also be useful in providing an easy upgrade path to some legacy OSs, if the OS already has a driver for the unaccelerated standard interface (like AHCI). There would be no need to provide OS install/upgrade support for the accelerated mode, since the standard interface could be used for these actions instead.

I think a more useful distinction between "true hardware RAID" and "fake RAID" would be: "can it process the RAID5/6 calculations independently from the main CPUs?"

If you read the specifications of the SmartArray B110i SATA RAID controller, you'll find it does not support RAID 5/6 modes at all, only RAID 0, 1 and 1+0 - which certainly makes it distinctly inferior to other SmartArray models. However, if you don't actually want to use RAID 5 or 6, then that's not a problem for you.

Granted, a true hardware RAID would provide other features that might be important - like auto-rebuild and seamless hot-plug support. But hot-plug is supported at the OS level by modern OSs, and implementing auto-rebuild would usually be just a small matter of scripting.

By the way, a true hardware RAID design can become a bottleneck if the system CPUs can do the job significantly faster than the dedicated hardware. Something like this has already happened; see the cases of Promise TX4 and SX4:

https://ata.wiki.kernel.org/index.php/SATA_RAID_FAQ

MK

[Matthew Darcy_2]

again, valid points.

the bottom line for me is to stay clear of the 3rd party modules, that's the main issue for me. I understand that a big array of modern day cpu's will wipe the floor with an onboard controller, and to that end I often use linux software raid with mdadm to great effect, I have zero issue with full software raid, I'm just not comfrotable with the middle ground of a hardware controller that requires external components to see the root disks, as these are not high IO boxes I should be fine with what I was earlier classing as true hardware raid.

the list in the first post is perfect as it will allow me to work out which machines off the support I want rather than make another incorrect purchase.

big thanks for the input, it's all useful.

[Jimmy Vance]

Just to add some definition. Some of the lower end DL server like the 100 series use an embedded RAID controller that most call "fake RAID", it's not really fake, but it's software RAID controlled by the driver, not hardware RAID like controllers that use the cciss driver.

To add to the confusion the current shipping 100 series systems now call the controller a Smart Array B110, but it's still software RAID controlled by the driver and again, it does not use the cciss driver.

Smart Array B110 = software RAID

Currently all other Smart Array controllers = Hardware RAID

[Matthew Darcy_2]

that's pretty much what I was trying to say in my first post, albeit badly.

I didn't want to suggest software raid as that would be the OS controlled raid, mdadm for example, and wanted to make the distinction between a true onboard raid controller, and a disk controller with raid capabilities with specific software.

Fakeraid wasn't meant as a derogatory meaning, it just seems to be the accepted name for this class of raid device.

[Matthew Darcy_2]

the information in the posts above has allowed me to work out the correct model servers and hardware raid controllers required to meet my projects needs.