How does GRAID work?

I don’t wanna know how many customers GRAID got through the exposure from that overly positive first LTT video…
(Even if they only offer a “contact us for a price” kind of product)

That server isn’t using the GRAID product though is it? I don’t remember them mentioning that it was going in there, specifically SuperMicro told them that SM had to put the server together.

1 Like

The video does not have a graid card; the hosts mentioned in the video, that they confirmed Wendell’s testing.
-That the graid cards do not check the data on read (not sure about patrol checks/scrubs) and Will serve corrupt/ faulty / false data.

(Fun fact, my phone autocorrected GRAID as fraud…)

3 Likes

So after watching @wendell outline this in the recent video the answer to the op topic is this

“It doesnt”

As I suspected

Those who do not understand ZFS will eventually reinvent it. After decades of dealing with the problems it fixes.

3 Likes

My favorite thing from that video has been the TechBros that have popped up and adamantly insisted their $MEGA enterprise-grade san that has a “Raid 5” and “Raid 6” gui has never had a problem despite petabytes of data in and out.
… but the documentation for those systems show they have extra parity fields they use for data integrity, and the software on the san does checksum validation using those extra data fields on every read.

7 Likes

Hey all,

New here, longtime viewer. Bought the GRAID SR-1010 after Linus’ video (yay), then quickly stumbled upon Wendell’s video (yikes).

It appears GRAID Tech recently released the SR-1010, upgrading their Nvidia T1000 to an A2000 which includes ECC memory, while also supporting PCIe 4.0. They’ve also released v1.2.2 of their software which touts automatic data correction, with an incredibly brief news release.

(Can’t post links; you’d have to Google search the below):

news-graid-announces-new-data-consistency-check-feature-set-delivering-critical-protection-against-data-corruption

I’ve reached out asking what exactly that automatic data correction software entails, while also referencing the video “Hardware RAID is Dead and is a Bad Idea in 2022” and if any of the concerns regarding bit rot were addressed within the SR-1010 on v1.2.2 software.

Would love to share what they report, as well as any questions here to forward to them.

With these changes GRAID at least reached the minimum to be taken seriously to maybe look into it further, thanks for sharing your experiences!

Can you say how much it cost (without the dGPU)?

It can’t address them. It’s as vulnerable to bitrot as software raid or hardware raid (without extra application checksums). The software sounds like a scrub function, so it can go through and find data errors.

AFAIK bitrot on HDDs (and SSDs?) isn’t totally bad. Apparently they store ECC on disk and can detect a massive 11 bits worth of errors - in a 4096 byte sector. If the ECC fails to verify the data, the disk returns a read error, and your software / hardware raid can restore the data from parity. (Though I’m having a hard time finding good sources for sector ECC on HDDs). ((And this is separate to the 520 byte data integrity fields Wendell talks about in his video)).

If you need really fast raid across multiple NVME drives, and want to save CPU parity calculations, the solution doesn’t seem all bad. Just keep regular backups.