basic raid settings

[mrstrong]

just setting up a new dell server as a hyper-v host.

It's Raid 10 with 4 x 2TB SSD drives

 

Are the defaults ok for read / write /stripe i.e. :

stripe element size: 256KB

read policy: read ahead

write policy: write back

[PotNoodleTech]

Yes that would work fine.

 

I'd probably use Raid5 though and leverage the extra 25% capaity

[3s-gtech]

I wouldn't use RAID 5 for a host - the read write speed is far worse than 10. Enjoy the lovely performance and redundancy, capacity is cheap.

 

Defaults are fine - I usually try to put the host itself onto independent storage too.

Edited March 3, 2023 by 3s-gtech

[Davit2005]

Have the host system use its own drives in a Raid 1, these don't need to be that big then use the RAID5 or RAID10 for the VMs and their associated disks. But if you are using SSDs I'd debate using RAID 10, maybe even create a RAID 5 with a hot spare.

 

Or even RAID 6.

Edited March 3, 2023 by Davit2005

[PotNoodleTech]

I wouldn't use RAID 5 for a host - the read speed is far worse than 10. Enjoy the lovely performance and redundancy, capacity is cheap.

 

Defaults are fine - I usually try to put the host itself onto independent storage too.

 

I guess it depends if the OP needs the space or not

 

I've never found that - Raid 5 has very good read speed especially if your using a decent raid controller with cache.

[3s-gtech]

Of course, I meant write speed. Rushed reply. RAID 5, at least traditionally, was a bad idea where you had lots of writes, and a host usually does especially if you have anything like a SQL server in there.

[mrstrong]

mmm, lots to think about here (as usual )

out of interest how large do you have your OS / C partition on the hyper-v host?

is 100GB enough ?

 

Like I said I was thinking of one virtual disk raid 10 (with my 4 ssds) then putting

on Server 2022 in a 100GB C: partition, created during the OS install

Then after OS installed create a new partition D: (using all remaining space) for the hyper-v VMs.

 

(edit not sure I have budget to buy any more disks right now)

Edited March 3, 2023 by mrstrong

[3s-gtech]

Can you put the host on physically separate disks? That's an ideal - again, not always possible.

 

100GB should be adequate for the OS. Should be plenty, but we've all used Windows!

[mrstrong]

Can you put the host on physically separate disks? That's an ideal - again, not always possible.y

100GB should be adequate for the OS. Should be plenty, but we've all used Windows!

 

so you mean ideally buy another 2 disks and create a new virtual disk (eg raid 1) for the host OS server 2022

then use my existing four as Raid 10 for the VMs ?

 

The Raid controller is a PERC H755, would that work e.g. can you have several virtual disks with different Raid levels

on same controller ?

 

If you did only have the 4 discs to play with how would you do it?

[3s-gtech]

If your backplane will take the extra disks, yes I'd do that. Looking at the specs it's fine - supporting multiple RAID arrays of different levels is part of the business of being a RAID controller.

 

With 4 disks max I'd do as you've suggested - a partition for the OS and another for the VMs. The advantages of a separate array for the OS are more speeeeeeed, reduced risk of VM disk failure through wear, and flexibility of OS disks.

[mrstrong]

If I go ahead with my plan of one raid 10 virtual disk OS on C:, VMs on D: would it be possible

to "migrate" OS to a new raid at a later date, i.e when I get some budget to buy another couple disks ?

 

separate question: could I just buy a couple of cheap ssds off amazon or would that void my dell hardware/support warranty

i.e. do I have to buy dell approved disks from dell ? I'm assuming I'm allowed to add disks without voiding warranty !

[3s-gtech]

Not that easy to migrate the OS in that case, but you can simply build a new hypervisor OS on the new RAID then attach the VM RAID array to it (which would include the old OS partition, which you could nuke).

 

You can use cheap SSDs, I wouldn't. I've used Enterprise class SSDs (not actually that expensive though) to do this. I can't tell you with certainty about the warranty - I don't believe it will void it but of course you wouldn't be covered on issues caused by them.

[msi_school]

Yes that would work fine.

 

I'd probably use Raid5 though and leverage the extra 25% capaity

 

Raid5 on HDDs is problematic with large disks, the unrecoverable bit error rate can be lower than the number of operations required to rebuild the array Raid10 and Raid6 put this off by reducing the number of operations required to rebuild the array.

 

Whilst SSDs use a different technology the underlaying issues with the number of write cycles still exists and there is not a lot of experience in the use of large SSDs and Raid out there. there are papers saying that SSDs reduce the URE problem by up to 100 times and others that are less positive.

 

Personally I would stick with the Raid10.

[PotNoodleTech]

Raid5 on HDDs is problematic with large disks, the unrecoverable bit error rate can be lower than the number of operations required to rebuild the array Raid10 and Raid6 put this off by reducing the number of operations required to rebuild the array.

 

Whilst SSDs use a different technology the underlaying issues with the number of write cycles still exists and there is not a lot of experience in the use of large SSDs and Raid out there. there are papers saying that SSDs reduce the URE problem by up to 100 times and others that are less positive.

 

Personally I would stick with the Raid10.

 

Agreed. I do use Raid 6 on the backup servers (3+tb drives) but Raid 5 on my Hyper V servers which all have 400-800GB SSD's.

 

That being said - I've never lost data on Raid 5 in 20+ years but I've always used good servers with good raid controllers. If your using 8tb drives on a £500 server using Software raid built into windows server you will no doubt have a much worse experience than me.

[localzuk]

For critical services, RAID10 is a good option. For backup servers, I don't mind RAID 6 these days. I don't use RAID 5 any more at all due to the issue @msi_school mentions.

 

Though, for our network, we're doing away with RAID entirely, and have moved to a HCI setup with Storage Spaces Direct. But that needs more than 1 host.

[PotNoodleTech]

The reason I use Raid 5 on SSDs is to reduce unneccasary writes and write speed isn't a problem on SSDs generally anyway.

[nobley]

I would second the comments about RAID 5. It has poor write performance compared to RAID 10. I would also suggest double checking your RAID controller supports SSD TRIM correctly. I've seen some (cheapo) controllers where TRIM didn't work in a RAID configuration and the SSD got slower and slower. Taking the drive out and running a manual TRIM on it fixed the problem (which then slowly got worse again in the original system).

[mrstrong]

just a quick follow-up on this, with 4 ssd in a raid 10 is there any benefit (e.g. performance?)

to having more than one virtual disk in the raid 10 setup?

 

e.g.

one virtual disk with partitions C: for OS and D: for VMs

vs

two virtual disks with first one for OS and second one for VMs ?

[Koldov]

This thread prompted me to have an in-depth look at my servers...

 

 

'Server 2' was our old original server, it was set up before my time and I always wondered (but now I know) why it was slower than 'Server 1' which I set up and moved most of the important stuff to (SIMS, File Share, etc.), it wasn't slow enough to really be a problem so never really bothered looking at it in any detail, just when working between the two though it was noticeable.

 

I don't have the need to change anything major like adding disks to change RAID levels because it would be a cost (both in time and money) for no real benefit as it doesn't have a heavy workload (it only hosts a couple of VMs, one for secondary DC, one for Print server/WSUS and one for a Windows 10 RD instance).

 

But interesting to see the difference in things like Write/Read and Cache policies and stripe size...

 

Can anyone explain (in basic terms) what the difference is and what's best practise for a basic 'do it all' school server?

Edited March 27, 2023 by Koldov

[mrstrong]

both servers raid look good to me e.g. you have OS on 2 seperate physical disks in raid 1

then the vms on different physical disks (raid10 on srv 1, raid1 on srv 2)

this was what was recommended earlier in thread.

 

My issue is I only have 4 disks and have set them up as raid 10 with one virtual disk, OS on C:, VMs on D:

I'm now wondering if it would have been better to create two virtual disks like you have.

Of course my two virtual disks would be on the same underlying 4 physical disks unlike yours.

[Koldov]

Yeah, I think this was the problem with whoever did the spec for the original server... they didn't put in enough disks for the DATA virtual disks to be anything other than RAID 1 really. I'm sure it's OK for redundancy, but it can't be great for write times (?). Or maybe they just wanted that 1 - 1 data security...?

 

I have to say I cheated a little on the newer server and copied the separate physical disks for OS/DATA idea from that (also because it's what I've been doing in personal machines for years ), but did a bit of research to make sure I had enough disks for a decent level of RAID on the DATA disks.

 

Didn't do enough research on the RAID controller though obviously as the PERC H310 is about as basic an entry level controller as you can get. It doesn't have any cache, so I think that's why it might have been set as 'write-through' (?). It must have come from DELL set up like that though because I only really looked at it much later on and never really got a handle on whether I should change it (still haven't and now I know the H310 has no cache it might be a moot point).

 

So you have:

 

4 DISKS > RAID 10 = 1 VIRTUAL DISK > 2 partitions (C: & D: ).

 

But you are thinking of:

 

4 DISKS > RAID 10 = 2 VIRTUAL DISKS > disk 1 partition (C: ) & disk 2 partition (D: )

 

Unfortunately I can't help you on the benefit (or not) of virtual disks on the same physical disks, I guess there's a lot of factors that come in to play. I can see where you are doubting the setup, but if the issue is writing to the physical disk (even through 2 separate virtual ones) then I'm not sure you can avoid it (?) or that it will make a difference.

 

To separate them out you could go single disk for the OS and the other 3 in RAID 5 for DATA (?)... If you were feeling brave... might be waste if you got 4 big disks though, so I can see the logic of lumping them all together and then dividing them up as you have.

Edited March 27, 2023 by Koldov

[mrstrong]

yes i think you are correct about the write back vs through i.e. without a cache on the controller

it wouldn't matter.

To get some real data you could try live optics (free I think) from dell to see how your server is performing etc:

 

https://www.dell.com/en-uk/dt/live-optics/index.htm

[mrstrong]

did some googling over lunch and nothing definitive, but get the feeling if you only have one disk

group like me (4 physical disks in raid 10) then stick with one virtual disk rather than slicing:

e.g. from this (old) thread:

https://www.dell.com/community/PowerEdge-HDD-SCSI-RAID/Create-1-large-or-2-smaller-virtual-disks/td-p/4062822

[PotNoodleTech]

Ideally run a separate disk group for the O/S e.g most modern dells etc can take 2 x SSD in the rear by the PCIE slots for the O/S.

 

If your server doesn't have that then one disk group.

[TwistedHelixis]

read policy: read ahead

write policy: write back

 

I am also just setting up a new Hyper-V host server like you, but just read the following in regards to the 'read ahead' policy and wondered what everyone thought?

 

Hypervisors are sometimes referred to as "IO blenders". They make it very hard for any algorithm to predict what they'll do next because they don't let each VM have its own real SCSI disk. Because of this, you definitely don't want straight up read-ahead.

Edited February 9, 2024 by TwistedHelixis