Power Loss Followthrough
-
I asked a question here a few weeks ago. Why do you need SSDs or RAID controllers that will continue writing after a power loss, and the answer was: in case the power supply goes.
That makes sense, but where does the progression stop?
I mean, if the power supply goes, doesn't the motherboard, etc., also go?
At what point does this all become moot as components that aren't protected by power loss circuitry get involved and fail?
-
It's particularly important for RAID controllers to have time to flush out their cache as even a very very minor hiccup in power can cause them to barf.
-
The issue is components that do caching. It's all about either not having storage do anything or getting confirmation that writes have completed and will persist.
-
@BRRABill said:
I mean, if the power supply goes, doesn't the motherboard, etc., also go?
At what point does this all become moot as components that aren't protected by power loss circuitry get involved and fail?
This isn't about continuing to process data. This is about writing data to the physical disks that have already been sent to the raid controller and is currently in volatile memory, if this is not copied to a disk before it looses power all of this data will be lost.
-
Data that systems believe has been recorded. Transactions that something thinks has been processed but hasn't finished, in actuality.
-
But if the power goes off (via power supply loss or whatever) wouldn't data being sent TO the controler/cache already be interrupted in mid stream?
-
@BRRABill said:
But if the power goes off (via power supply loss or whatever) wouldn't data being sent TO the controler/cache already be interrupted in mid stream?
Data sent but not received isn't an issue as nothing believes that it has arrived yet. It is the data that has arrived at the controlled but has not gone to disk yet that is the issue.
-
@scottalanmiller said:
@BRRABill said:
But if the power goes off (via power supply loss or whatever) wouldn't data being sent TO the controler/cache already be interrupted in mid stream?
Data sent but not received isn't an issue as nothing believes that it has arrived yet. It is the data that has arrived at the controlled but has not gone to disk yet that is the issue.
I thought this was why some raid cards have their own batteries?
-
@dafyre said:
I thought this was why some raid cards have their own batteries?
Exactly. That what we are explaining the need for
-
So once the system (aka the motherboard) sends it to the controller, it assumes it has been written.
So if it never hits the controller, it's like it never happened?
-
P.S. I am assuming the H710 (512MB) has a battery?
-
Should be flash backed and not volatile.
-
@BRRABill said:
So if it never hits the controller, it's like it never happened?
Right, the controller is the first component that reports back up the stack that the "data has been written to disk." Things up the stack need to be able to trust that report.
-
I want to add something to this,
I'm assuming that the battery backup or non volatile Flash don't write anything while the power is out. because, the RAID controller battery doesn't have enough power to keep the drives spinning, and obviously the non volatile Flash has no power at all.
Instead, these backups keep the RAID data alive until the system comes back online and then finishes writing the data to disk.Right?
-
@Dashrender Correct.
-
that's what I figured.
Thanks
-
Yes, they just maintain "stasis" until the power comes back on.
-
Where does the "power loss circuitry" in an enterpise class SSD fall into this?
-
@BRRABill said:
Where does the "power loss circuitry" in an enterpise class SSD fall into this?
If I were to guess, I'd say the same as the RAID controllers. The data comes in and is written to some non volatile place, but not reported as finished being written to the RAID controller until it's done writing to the final destination on the SSD.
-
@Dashrender said:
@BRRABill said:
Where does the "power loss circuitry" in an enterpise class SSD fall into this?
If I were to guess, I'd say the same as the RAID controllers. The data comes in and is written to some non volatile place, but not reported as finished being written to the RAID controller until it's done writing to the final destination on the SSD.
They have capacitors (super caps?) in them. These act like a battery.
