@MrPete
I mean doesn't help for performances.
indeed UPS is always required to prevent data loss on power loss.
Last edited:
I mean doesn't help for performances.
indeed UPS is always required to prevent data loss on power loss.
I am not an expert on this topic, but I think you guys are mixing a lot of stuff together.
So there are two things affected here.
First one is the ability to write data after a power loss. Because of that ability, the drive can way faster signal to the system, that a sync write is finished (which is not (yet)). So this is not only a few tens of ms like you describe, this could be some sync writes that would take seconds but now take only ms instead. That is why a on paper "slower" Samsung PM953 will easily outperform a "fast" Samsung 960 Pro in that use case!
Even though the 960 Pro as tripple the sequential write speeds and more than twice the random write 4KB, QD1 performance.
The second thing is security. My understanding of ZFS is, that this does not apply to ZFS because of checksums but I could be very wrong on that one.
Their reasoning was, since UPS are so cheap, better save than sorry.
PLP is so that in case of a system halt, the drive can safely flush data in-flight (or data that resides in the drive’s DRAM or SRAM cache buffers) to the persistent or non-volatile Flash memory and maintain the integrity of the SSD’s mapping table so that the SSD is recognized and useable again upon reboot of the system.
So there are two things affected here.
First one is the ability to write data after a power loss. Because of that ability, the drive can way faster signal to the system, that a sync write is finished (which is not (yet)). So this is not only a few tens of ms like you describe, this could be some sync writes that would take seconds but now take only ms instead. That is why a on paper "slower" Samsung PM953 will easily outperform a "fast" Samsung 960 Pro in that use case!
Even though the 960 Pro as tripple the sequential write speeds and more than twice the random write 4KB, QD1 performance.
The second thing is security. My understanding of ZFS is, that this does not apply to ZFS because of checksums but I could be very wrong on that one.
UPS only protects you from a power loss, but there are many other reasons why your system could halt. We had a huge discussion in TrueNAS community about why TrueNAS recommends a UPS, when there is no actual technical reason for this. All the power loss problems apply to traditional FSs but not to ZFS.
Their reasoning was, since UPS are so cheap, better save than sorry.
Last edited:
And neither does PLP. To quote the OpenZFS page I linked above (emphasis mine):
Please re-read the OpenZFS page I linked (and the Kingston further details if interested. Not all vendors are that forthcoming with their methods!) Your explanation dates to much older hardware. For almost a decade, firmware and hardware power protection have been equivalent, for quality drives***. Of course on-board RAM cache helps speed things along, but that's actually a separate topic.
Of course
Interesting. I guess TrueNAS community considers loss of unwritten async data inconsquential. To each their own.
*** Of course ANY drive with any technology can have firmware bugs, HW failure etc. Thus, even though all drives supposedly are "safe" today, various testers have discovered that's not always true. Buyer Beware is always good advice.