VROC RAID 5 journaling drive sizing

[Erik_P]

I'm putting together a new system that will include a VROC RAID 5 array for file storage. I've done my best to get up to speed on requirements for closing the RAID write hole, but I can't find a document that clearly illustrates the techniques and requirements.

Specifically, how big should the journaling drive be? I'd be surprised if it needed to be the same size as the array members, but I can't find a sizing guide anywhere. I'd appreciate it if someone could point me in the direction of an authoritative document or sizing formula.

Best,

Erik P

[BrusC_Intel]

Hello, Erik_P.

Thank you for contacting the Intel Community Support.

I checked your ticket regarding Intel VROC and RWH, I will be glad to assist you.

Please allow me to verify if there are any specifications or requirements for this.

in the meantime, I would like to know if you are planning on implementing VROC on an Intel server board or a different EOM system, also if you are aware of the VROC version that will be used.

I will contact you back as soon as possible.

Best regards,

Bruce C.

Intel Customer Support Technician

A Contingent Worker at Intel

[Erik_P]

Hi Bruce,

Thanks. System specs will be as follows:

Supermicro X11DPH-I mainboard w/ 2x Xeon Silver 4215R, 256GB ECC RAM installed

VROCPREMMOD is the part # for the VROC license dongle

RAID 5 array will be 6x Samsung PM983 m.2 NVMe drives (part # MZ1LB3T8HMLA)

•  2x "ASUS Hyper M.2 X16 PCIe 3.0 X4 v2" cards will have 3 drives each
•  The Supermicro board supports appropriate bifurcation of the PCIe lanes on the x16 slots
•  BIOS does support for spanning across VMD controllers  

Nothing else special in this machine... the boot/OS drive is a separate RAID 1 array using the 2 onboard m.2 connectors (not spanned).

Yours,

Erik P

[BrusC_Intel]

Hello, @Erik_P.

Thank you very much for waiting.

After reviewing your details and the available documentation, I can confirm the following, but please keep in mind that you should also check with motherboard/system manufacturer for details on how they implement RAID-5 write hole protection, this is because the information we can provide is general and may vary depending on the system manufacturer or they may have other information.

1. The "journaling drive" needs to be at least as big as the smallest drive in the RAID volume (it will dictate the maximum size of the RAID volume).

2. For RWH protection, we recommend to work on "Distributed mode" (where the journaling information is stored across all 3 RAID members) and not "Journaling Drive" (where journaling information is stored on a single drive).

If I can help you with anything else, please let me know.

Best regards,

Bruce C.

Intel Customer Support Technician

A Contingent Worker at Intel

[Erik_P]

Hi Bruce,

I'm confused. What you describe sounds like parity striping, which is already a part of the RAID 5 scheme. Is Intel referring to this when using the term "journaling"? I had interpreted the journaling behavior as something different. If the journaling drive is really just a re-location of the parity data, then I have to say this is a poorly thought-out solution. Distributing parity information across all member drives is the kernel for RAID 5 fault tolerance. Placing it all at a single point of failure just isn't smart, and I can't believe this is what's being described. A detailed description for how Intel VROC implements RAID write-hole closure is very much needed (and no, this is not dependent on motherboard--it's all done by the VMDs built into the Intel processors).

Thanks,

Erik P

[BrusC_Intel · Accepted solution]

Hello, @Erik_P.

Thank you very much for waiting

Please review the following details and let me know if you have any questions.

Journaling is different than parity. Journaling happens first, even before the data is committed to the RAID5 volume, and once the RAID 5 volume is ready, the journal is no longer needed.

The reason why it was mentioned that the journaling drive should be at least as large as the smallest drive member in RAID5 is owing to endurance considerations (in case of dedicated journaling), while journaling does not actually require so much space, it does need to be written over again and again, requiring good endurance/capacity, and is what could make distributed journaling more effective.

Distributed journaling utilizes the PLI functioning of the drives, which is in turn using memory on the drive to store data instead of NAND, making endurance not a factor or concern anymore for journaling.

The other option is to use an Intel® Optane™ SSD as a dedicated journaling device, which much lesser capacity drive with way higher endurance (as a dedicated journaling solution).

Best regards,

Bruce C.

Intel Customer Support Technician

A Contingent Worker at Intel

[Erik_P]

Thanks Bruce, very helpful. A smaller Optane drive it is...

Yours,

Erik P

[BrusC_Intel]

Hello, Erik_P.

Perfect, I'm glad we can be of assistance.

I will proceed to close the thread right now, if you require further assistance, please open a new thread as this one will no longer be monitored.

Best regards,

Bruce C.

Intel Customer Support Technician

A Contingent Worker at Intel