Is there a place to be RAID10 from mdadm and LVM2?

Question

Iron

Who faced / collected: how does an array created in Linux using mdadm and LVM2 behave?
I saw recommendations for such an implementation in the redhat docs.
Those. we take 4 disks and assemble two mirrors using mdadm, and then using LVM2 we create a single volume on top of these arrays. On structure - RAID10, but on implementation not absolutely it.
How will it work in HDD and SSD drive options?
Are there any advantages and what are the disadvantages?
For what tasks will this be a good solution, and for what - a bad one?

Answer 1

7For example, this way https://www.allcloud.io/how-to/create-software-rai... we collect raid10, and on top of it lvm.
I have not tried with raid10 myself, the mirror works great.
The disadvantages are obvious - software raid, hardware raid is faster and more reliable. But this one is free. Synchronizes for a long time with the distraction of processor resources.
If you want to cache with ssd, then this is IMHO additional software like https://www.rath.org/ssd-caching-under-linux.html.
I use proxmox virtualization hosts.

Answer 2

> I.e. we take 4 disks and assemble two mirrors using mdadm, and then using LVM2 we create a single volume on top of these arrays. On structure - RAID10, but on implementation not absolutely it.
Bad idea. Precisely due to the fact that you are not doing a raid10, but a bicycle.
The whole ideology of RAID was built not around data backup, but around minimizing downtime and speeding up work.
You are doing lvm2 over the ride, that is, in case of failure, you will need to manually restore - and this is the opposite of what raid offers - take all the time spent on yourself, and you just need to re-plug the broken screw for a whole one.
It's better to make raid10 regularly (by means of the OS, by means of the controller) and on top of it already stir up lvm2 in order to work with dynamic partitions.

Answer 3

Is there a place to be RAID10 from mdadm and LVM2

Yes.

How will it work in HDD and SSD drive options?

It will work fine, both HDD and SSD. The only question is why do you need it, what tasks should it solve. We must proceed from this.
So it doesn't always make sense.
It is possible to make the tenth raid from an SSD, but situations do not often arise when it makes sense.
It's usually easier to buy a faster SSD.
In general, there are many subtleties in creating raids from SSDs, and the reason for this is simple.
HDD is essentially a dumb piece of iron, it does only what it is told no logic (although I must say lately and HDD have become wiser, take the same NSQ)
But an SSD is essentially a computer with a powerful processor, a decent amount of RAM, the firmware that controls it all - well, and an array of flash memory itself. And this computer has its own logic laid down by the manufacturer, and there are many options, it is difficult to find two identical SSDs in all this diversity.
Therefore, it is difficult to clearly predict how exactly which SSD will behave in a particular situation.

Answer 4

Who faced / collected: how does an array created in Linux using mdadm and LVM2 behave?
I saw recommendations for such an implementation in the redhat docs.

It is possible to look at such docks RH, where they give bad advice?
In LVM2 RAID0, there is only one plus, which is creating snapshots and adding new HDDs.
Only now such a solution will read and write sequentially and in general there will be no increase in speed.
The second point is that the reliability in case of failure of such a scheme is debatable. In fact, you can fly out 1 disk or 2 disks from different arrays. restoring such a layer cake is not very convenient, in case of problems - restoration can be a quest. Therefore, it is easier to do everything on mdadm at once, for example.
Yes, as usual. The difference is minimal.
There are no answers and never will be. Everything is decided by benchmarks and from the tasks themselves. At the same time, benchmarks are purely for your needs and for your hardware.
But I would generally leave both mdadm and lvm and already look at ZFS with its goodies. There you already have LVM, and Raid, and snapshots and caching, etc., etc. In the coming time, the FS will dominate. Two levels of cache can be configured. Memory-\u003e SSD- and everything is already on the HDD.