What kind of Raid to build from SSD disks in the server?

Question

RAID

Colleagues, help out. There is an HP DL580G7 server with 8 drive bays. SSD drives are inserted into these compartments, also HP, i.e. All 8 bays are occupied by SSD drives.

The question is, which RAID is better to assemble to ensure maximum performance of the disk subsystem from SSD drives? But at the same time, fault tolerance should be at a sufficiently high level.

And in general, does it make sense to use all SSD drives? Maybe leave some of them in stock, and build a raid from a smaller number of disks?

Upd:

I'm clarifying . Server HP 580 G7 ( 584086-421 )
SSD drive ( 572071-B21 ) (neat pdf)
Controller: HP Smart Array P812/1G ( 487204-B21 )

Answer 1

Why not make RAID 1+0? Those. 4 pieces of RAID 1 are collected in RAID 0.

Answer 2

I think the 10th raid will be the most it ... although of course you lose a lot of disk space

Answer 3

raid 60 - more reliable (I would still choose this option)
raid 10 - faster (in this case, it mostly depends on the performance of the controller)
redundancy losses will be the same in terms of disk space.
about working with disks.
1. if you have analogues of Intel X25-E (I suspect that this is most likely the case), then they can be installed on any controller without problems and marked out completely.
2. if you have something on MLC chips and not the latest generation (vertex 3 comrades), then either use them with a raid controller and an axis that can trim, or not use them at all =)
3. if MLC, but fresh, then you can use it without trim, but it is highly desirable to mark up not the entire volume of drives. leaving 10% unallocated - double the service life. I would leave ~ 50%, which is without trim, what is with it.

Answer 4

What controller do you have? Does it work with ssd drives? If not, then most likely he will quickly take them away.

Answer 5

What kind of performance are you planning to get from SSD drives? Because a lot depends on the load profile. On write operations with a large block, you may not see much performance. Also, do not forget about the bottleneck - the raid controller itself, which may not digest all the performance.

Answer 6

The controller holds sata 3 Gb and disks work on sata 3Gb
So on doubts whether the controller will pull out - will pull out. Bus pci2 4.6 Gb in both directions. There disks are only 60 GB, it is unlikely that infa will go back and forth, there is more miscalculation on the iops of drives (unless, of course, you stir up video editing there).
As for compatibility with ssd - I doubt that the controller is suitable, nevertheless, you need to think about something else - when the speed on the sata is cut in this way, and the issue of the ssd resource is also solved, then they choose the option of disabling one of the cache mechanisms on the controller or ssd disk.
And for sure, I would put a couple of screws in the Sasovsky raid for backups, they will have a speed of 6Gb, and yet, if their heads don’t rattle, there is a chance that the drive controller won’t burn out like it happens with ssd, and a limited resource of cells like there is none.