NVMe Microsoft and FUA (force unit access)

[Redstorm]

Their have been some reports around of SM951's performing poorly in write test scenarios, As it turns out the microsoft driver is issuing FUA IO commands to the SM951, Why? the answer is quite simple. The SM951 does not have a battery backed cache, so it cannot guarantee writes in a power loss situation.

This is not the case with the Intel 750 as they have full power loss protection and hence FUA IO commands are not required.

If you can disable FUA and you have a power loss situation when it is writing you will lose or corrupt data most likely.

See here for the spec, explains why it is happening.

https://msdn.microsoft.com/en-us/library/windows/hardware/jj134356.aspx
Search the page for "FUA" to find the relevant section.

WRITE(10). Support for force unit access (FUA) is mandatory for individual physical disk
 drives or RAID controllers that contain volatile (non-battery-backed) cache memory and 
must cause the data sent with this command to be committed to physical media before the
 command completes.

I have been keen to get a SM951 NVMe model when they become available but this situation has me leaning back towards a Intel 750, unless power loss protection can be added to the m.2 SM951's (possibly not as capacitors would likely protrude too much and cause the device to exceed the spec on size).

 

[RamCity]

Their have been some reports around of SM951's performing poorly in write test scenarios, As it turns out the microsoft driver is issuing FUA IO commands to the SM951, Why? the answer is quite simple. The SM951 does not have a battery backed cache, so it cannot guarantee writes in a power loss situation.

This is not the case with the Intel 750 as they have full power loss protection and hence FUA IO commands are not required.

If you can disable FUA and you have a power loss situation when it is writing you will lose or corrupt data most likely.

See here for the spec, explains why it is happening.

https://msdn.microsoft.com/en-us/library/windows/hardware/jj134356.aspx
Search the page for "FUA" to find the relevant section.

WRITE(10). Support for force unit access (FUA) is mandatory for individual physical disk
 drives or RAID controllers that contain volatile (non-battery-backed) cache memory and 
must cause the data sent with this command to be committed to physical media before the
 command completes.

I have been keen to get a SM951 NVMe model when they become available but this situation has me leaning back towards a Intel 750, unless power loss protection can be added to the m.2 SM951's (possibly not as capacitors would likely protrude too much and cause the device to exceed the spec on size).

I think it's only a matter of time before someone gets creative enough to adjust the ID's of the NVMe SM951 drive in the windows registry and use the Intel drivers instead.

In the mean time, there is also the AHCI version of the SM951 which stacks up very well against the Intel 750 and doesn't have the FUA issue.

Rod

 

[Redstorm]

I think it's only a matter of time before someone gets creative enough to adjust the ID's of the NVMe SM951 drive in the windows registry and use the Intel drivers instead.

In the mean time, there is also the AHCI version of the SM951 which stacks up very well against the Intel 750 and doesn't have the FUA issue.

Rod

FUA is their for a very good reason, to protect against devices with volatile caches without power loss protection. As far as im aware both the AHCI and NVMe versions don't have power loss protection.

Hacking the registry to get the SM951 to use the intel driver (which is designed for a device with power loss protection) is not a very smart thing to do. This will lead to corruption and data loss.

I would like to see if Samsung can include a capacitor large enough to allow the SM951 to flush the volatile cache in the event of loss of power. But in the M.2 form factor i don't think their will be enough room to implement it.

However if you run a SM951 on a add in PCIe adapter their is plenty of real estate to implement power loss protection on the adapter card instead. This is probably the exact approach HP have taken with the HP Z Turbo Drive G2, which is powered by Samsung NVMe tech.

http://www8.hp.com/us/en/campaigns/workstations/z-turbo-drive-g2.html

Nothing on the HP web site about power loss protection for the Z Turbo G2 though.

 

[RamCity]

FUA is their for a very good reason, to protect against devices with volatile caches without power loss protection. As far as im aware both the AHCI and NVMe versions don't have power loss protection.

I would like to see if Samsung can include a capacitor large enough to allow the SM951 to flush the volatile cache in the event of loss of power. But in the M.2 form factor i don't think their will be enough room to implement it.

Couldn't a relatively inexpensive UPS be good insurance against power loss/brownouts/spikes?

 

[Redstorm]

Looks like its quite easy to enable and disable with no nasty registry hacks.