Recommended storage configuration

[timofcourse]

Im just setting up our Veeam B&R configuration and looking for a little guidance on how the backup storage should be connected to the Veeam server. We're running the Veeam backup server on a Windows 2008 R2 VM. This VM is running on a server connected to a NAS. On that NAS we have an array for our VMs and a separate array we plan to use for backup storage.

What Im wondering is should we configure the backup storage array as an ESX datastore with a virtual disk attached to the Veeam backup server or should we configure to connect directly to the Windows Server running Veeam, circumventing setting it up vSphere?

Thanks in advance!

[joergr]

You could also ask "I just saw two fridges in a store today. which one should i buy?"

OK, all jokin aside, IT DEPENDS. You have to provide much more details about your setup.

a) What kind of Server? Network Connection? Disks Setup? Raid Type in Server? Raid Controller in Server?

b) NAS Type? Vendor? Speed? Disks? Raid Type? Network (what is it, NFS, ISCSI, ...)?

Best regards,
Joerg

[timofcourse]

Ha.. fair enough.

VMware hosts are Dell PE R610s. 2 are connected to a Dell PowerVault MD3000i SAN via iSCSI and a 3rd server connected via ISCSI to PowerVault MD3220i all of which are routed through a dedicated Dell managed switch with jumbo frames enabled.

The first SAN (MD3000i) has all 15K SAS drives in a RAID 5 array with 2 LUNS - one for a datastore for each of the 2 Dell servers connected. The 2nd SAN (MD3220i) has 2 arrays - 1 with 15K SAS drives in RAID 5 for a datastore for the 3rd Dell server and then another with 3, 7.2K SAS drives in RAID 5. Its this final array we plan to use for backups.

Because of our small setup, I have Veeam setup on a Windows 2008 R2 VM on the 3rd host and it will also serve as the proxy as well.

Any other info that may be helpful?

Thanks a bunch!

[joergr]

OK. So if you HAVE TO use a virtualized VEEAM Server i´d recommend to attach a huge Volume via iSCSI using in-guest MS software iscsi initiator to it and format it GPT Style if >= 2TB. the MDx Series should do the job, also with SATA drives. Have you by any chance thought about going one step further (upgrading to equallogic/emc/netapp, if you want to stay with dell: equallogic)?

Best regards,
Joerg

[J1mbo]

First, those RAID-5 volumes are way more dangerous than might be appreciated. For example EqualLogic doesn't even provide RAID-5 as an option any more (except in the CLI).

But re the question, a 3-drive RAID-5 volume will give very poor random write performance (less than a single drive), so Veeam transform processing with be very time consuming. Depending on your data sizes, this may or may not be acceptable. To give some indication of the intensity of that process, a 6-drive SATA RAID-10 volume takes 29 hours to transform 1.3TB with five deltas.

More and more I tend to just use RAID-10 for everything - here for example it will cost just one SATA drive, yet offer double or more random throughput

[joergr]

Agreed. Thus, i was asking for a little upgrade to let´s say EQL. And yes, it is true, since FW 6 EPA Equallogic do no longer recommend using Raid 5. I personally use RAID10 in my production storage and raid 50 in my replica groups.

Another reason i like RAID 10 that much is the good rebuild time and very little performance impact in case of a failed drive compared to raid5,50 and especially 6.

[kjc3303]

Hi Joerg

firstly sorry to jump in on the thread but didn't think my question warranted a new thread.

OK. So if you HAVE TO use a virtualized VEEAM Server i´d recommend to attach a huge Volume via iSCSI using in-guest MS software iscsi initiator to it and format it GPT Style

I have a question re your comment above as I have configured a different way for my Veeam VM

I have connected my LUN to my EXSI host via ISCSI (rather than using in-guest MS software iscsi) and created a large vmdk for my backup repository on this datastore. My Repository is only 2TB.

Is there anything wrong with my config? Everything is working but now I am concerned that I may be losing performance\not following best practice?

Thanks

Kevin

[J1mbo]

Only that there is a 2TB limit and mounting it is dependent on vmware (and the VM) being working.

[joergr]

There is nothing wrong with your config, and there are no general best recommendations because it depends on your setup and hardware. The only main problem i see with your config would be the need to at some point in time upgrade to a store > 2TB, which VMware can´t handle by now. Only if you would use RDM Disk. But if you can stick with 2TB max, everything should be all right. Another way would be to create a second vdisk with 2TB if storage goes up the 2TB barrier.

In general, ALL connectivity which do NOT throttle data traffic (this is iSCSI, vDisks, NFS, FC, etc) would be alright. Only NTP or CIFS i would NOT use for backup targets. Don´t mix this with transportation mode, NTP can do great if you can afford 10 GB infrastructure

[kjc3303]

Thanks

[J1mbo]

There is nothing wrong with your config, and there are no general best recommendations because it depends on your setup and hardware. The only main problem i see with your config would be the need to at some point in time upgrade to a store > 2TB, which VMware can´t handle by now.

You can always present multiple 2TB disks to the guest then use one of the varieties of Windows striping to amalgamate them. This works well with physical drives (multiple 2TB mirrors to get nTB RAID-10 without BIOS support), would need some thought in the effective RAID level being built in the virtual space of course.

[joergr]

correct, but before doing that i definitely would expose a huge xTB iSCSI LUN to the guest via software iSCSI initiator and place a vmxnet3 adapter inside the guest, then format the LUN with GPT. Striping inside the guest should only be used when there is no other option (my opinion).

Best regards,
Joerg

[J1mbo]

I agree with that.

[dellock6]

Uhm, there has been another thread about these kind of configuration... I do not totally agree with this design, it adds an extra layer and lowers the RTO since it also needs an ESXi server to work correctly.

Luca.

[joergr]

Hi Luca,

I wrote "If you HAVE TO use a virtualized VEEAM Server....". Thus, given this special condition i personally see no better way. But if you have an idea, please share it.

Best regards,
Joerg

[dellock6]

Hey, didn't want to be rude
Simply put, we also used sometimes virtualized Veeam servers, both proxies (that's easy, ok) and also some repositories, but we then used iscsi initiator in the guest mounting directly a storage LUN, formatted via NTFS. If something goes wrong and you loose VMware, you can connect via another physical server (even a pc if you want) via its iscsi initiator to the lun, mount ntfs and start recovering.
When you grow in terms of VMs/TB to be saved, I think better solution is to have physical repositories...

As always, it's not perfect anyway; like all design considerations, there are many pros and many cons in every design...

Luca.

[joergr]

No no no Luca, you are absolutely right
I personally prefer physical veeam servers as well (besides my virtual proxies ofcourse). It´s just i had to find a way for someone who HAS to use a virtualized one

Best regards,
Joerg

[dellock6]

What I find hard to justify, from a designer perspective, is the fact of "having to"

Lack of budget for physical servers maybe?

Luca.

[TwystedPair]

Does it make sense to have a virtual Veeam server and a physical proxy or should it be the other way around?

[Gostev]

Depends on the storage. For example, direct SAN processing mode is only possible through physical proxy in case of FC SAN.