Storage with power loss protection (PLP) or not?

[yoggi]

Hi forum members,

Being new to Proxmox, I've been trying to read up on what is "the right way".

I hope you here with more experience have the patience to answer my questions.

I've been trying to read up on storage. However, the more I read, the more confused I become.

I have no plans to use my machine as a NAS, more like a regular computer with virtual machines and maybe some LXC containers. I also don't look for 100% up-time.

My question concerns storage (SSD) with power loss protection (PLP) or not. Some (e.g. here on the forum) write that PLP is absolutely necessary and others seem to think that it is not needed.

When can I get by with "regular" SSD disks?

What do I lose if I don't have disks with PLP? When is it justified to have PLP?

And what about old fashion hdd, do they have the equivalent PLP (do to spinning platters in any way)?

I hope you can take the time to answer my novice questions

yoggi

 

[_gabriel]

ZFS require DataCenters grades Flash disks for their endurance (many TBW) and their PLP for performance.
HDD haven't PLP, that's why Write Cache Battery Backed exist on HW Raid Controller.

Regular SSD is OK with regular EXT4/LVM-Thin , but like regular computers, if there is crash or power loss at the wrong time, data loss can corrupt somes importants files. Perfomances aren't guaranted with regulars SSD, because speeds announced are burst speeds, after their cache is full, write speed is slow.

 

[gfngfn256]

In a home lab environment & non 24/7 running, as long as you maintain regular backups on another medium of all your VMs & LXCs + detailed notes of everything you configure you should be good to go with even consumer-grade disks. Try to make as little change as possible to host PVE, so even if a re-install is necessary you'll be good to go in no time, so put as much as possible in LXCs/VMs - not within the host. With this you will also maintain a much more robust host that should run peacefully.

 

[alexskysilk]

ZFS require DataCenters grades Flash disks for their endurance (many TBW) and their PLP for performance.

it doesnt REQUIRE it- the use can BENEFIT from it.

HDD haven't PLP, that's why Write Cache Battery Backed exist on HW Raid Controller.

Not related. A RAID controller BBU is to account for controller write cache, not the drives.

but like regular computers, if there is crash or power loss at the wrong time, data loss can corrupt somes importants files.

PLP doesnt protect you from unwritten changes. PLP protects you from data accepted by the drive to be written, but remains in drive memory as a deferred write. SSDs do this because they cannot write to partial sectors the way hdds do.

 

[_gabriel]

Not related. A RAID controller BBU is to account for controller write cache, not the drives.

Sorry for my fast reply and bad wording.
My point was BBU and PLP are for performance,
they can guarantee their cache will be written to the disk, so writes and FSYNC operations are fast and secure.

 

[Dunuin]

In short:
If you need good sync write performance (like running a DB or using something like ZFS that does lots of those) or you are writing much (either big amounts of data like DVR/torrenting) or you got a high write amplification (doing lots of small random writes, nested filesystems, CoW filesystems like ZFS, mixed blocksizes, virtualization, ...) or you are continuously writing and not only in small bursts you want a SSD that got a PLP and high DWPD/TBW.

And yes, those SSDs cost more. But on the long run those expensive enterprise SSD actually might be the cheaper deal, because you don't need to replace them as often. Better to buy an Enterprise SSD once for 400€ than replacing a 200€ consumer SSD multiple times.

 

[yoggi]

Hi all,

Thanks for chiming in, I really appreciate you taking the time to answer my questions!
I now have few follow-up questions.

If I decide to start with a mix of SDDs, e.g. 2x Kingston DC600M with power loss protection (PLP) and 2x normal consumer SSDs (both pairs configured as RAID1 “mirroring”) where would it make the most sense to place the SDDs with PLP (boot or storage of VM etc.)?

I'm thinking that the boot disk with PVE would be the obvious choice for the PLP disks (I'm just guessing here).

I also have an old Intel SSD 80GB 320 Series Intern SATA 2.5" (SSDSA2CW080G310) that has not seen much use, would i make sense to use it for Proxmox logging (redirecting the log files to it instead of the boot disk in order to minimize wear on the boot disks)?

Thanks again
yoggi

 

[Dunuin]

If I decide to start with a mix of SDDs, e.g. 2x Kingston DC600M with power loss protection (PLP) and 2x normal consumer SSDs (both pairs configured as RAID1 “mirroring”) where would it make the most sense to place the SDDs with PLP (boot or storage of VM etc.)?

Boot disks don't need performance (even a HDD would do the job). And writes aren't that bad (too bad for pen drives/SD cards but a bigger consumer SSD would do the job) as long as you use a single disk and not ZFS for a mirror. But when using ZFS to mirror your boot disks I would still get Enterprise SSDs because the CoW of ZFS is killing those quite fast (my homeservers running ZFS as bootdrives already killed 4x 120GB TLC consumer SSD...not that bad because no data-loss thanks to raid1...but still annoying to pay for replacement drives and to disassemble the rack to replace them).

Highest wear and performance requirement you will have on the VM storage, so if you can't use Enterprise SSD everywhere, I would use them for that.

 

[alexskysilk]

I now have few follow-up questions.

I know that home use is usually of less criticality than one that is used for "business" production, but when choosing equipment for a given purpose, your questions should look more like this:
1. How much data do my applications generate/require?
2. Is the data all of the same value? is there some of it I can live without if its lost?
3. For the high value data, how critical is it that all writes are accurate at all time? if there is some leeway, how many seconds of potential writes lost are acceptable in case of failure (equipment, environment)
4. How much availability time loss is acceptable?

Any of your options are workable without some definitions of acceptance- what you ask isn't really answerable.