Several questions

[matteu]

Hello,

I have some general questions I would like some details if possible

1) I'm installing several veeam infrastructure and very often architect ask me to use 1 proxy / host.
I see this is a recommended for hot add mode but I don't find anything about NBD. I often use NBD on 10GB network and I can't use 4 task or more on each proxy because there is only 1 repository and he can't handle 20 or more proxy tasks.
I would like to know how do you manage it ? Is it good 1 proxy / host on NBD mode ? How do you do to manage task with proxy and repository ?


2) I'll have to backup to apollo xfs (SOBR with 2 200TB repo) and immutability the vmware infrastructure and I have NAS backup to (arround 80To).
I saw immutability is not possible for NAS backup but I suppose I can use it as repository target but data will not be protected right ?
If I want to use archive for NAS, I see I need to create repository because SOBR can't do this.

3) What happen if I use GFS on backup job and I have backup copy job too for this job ? My GFS points will be copied to the backup copy repository and that means all the full backup will be sent ?

I will have other question probably but I don't have them in mind now :p

Thanks for your help

[HannesK]

Hello,

1)

use 1 proxy / host.

just to be clear... are you talking about one HotAdd proxy per ESXi host? I only remember we had that recommendation some time ago for VSAN. But not as a general recommendation. I like the "all-in-one model" with one physical backup server hosting all roles up to 1000 VMs. With more than 1000 VMs, I would start distributing load over multiple servers. NBD is fine - we have many large customers doing that. For full-VM restore, a HotAdd proxy is a recommendation as it is often the fastest way to restore.

proxy tasks and repository tasks should be the same. otherwise you get a message like "waiting for resources: repository"

2) correct. NAS data can be stored on Hardened Repository. Backups are mutable in V11 though.

What is your idea about using the archive repository for NAS backups? Keep in mind, that it only moves data that has already X versions or got deleted on the source. Object storage and standalone repositories are supported. No SOBR

3) The backup copy job has independent retention. Per default, the copy process is "incremental forever". Only changes will be transferred.

Best regards,
Hannes

[matteu]

1) Yes, we never implement architecture with 1 proxy. We always choose 1 proxy / esx host (if proxy uses hotadd or nbd mode).
I saw it's good for vsan but other hyperconverged storage too.

I understand total proxy should be equal to repository task number but what I would like to know if how do you choose it ?
If we have 3 proxy and 1 repository. Do you try 2 task per proxy and 6 for repository and if there is no bottleneck during backup job you try with 3 task / proxy and 9 for repository ?

2) When you use NAS backup, is it better to dedicate an other proxy ? I suppose for 80TB data it's maybe better to have physical proxy with better hardware than juste virtual proxy with 8 cpu (to avoid costop on esx cpu).

3) Thanks I ignored it will not copy GFS points on backup copy storage !

4)When storage array is used, what do you prefer ? Use storage snapshot (on compatible hardware) or nbd / hotadd for virtual proxy ?
I will have to backup on netapp storage array with NFS datastore and I suppose it's better to use storage snapshot or direct NFS ?

5)When you design backup architecture, do you use dedicated network ?
If yes, how do you do exactly ? VBR + all proxy + all repository are on the same VLAN and all have only 1 network card ?

Thanks for your answer

[HannesK]

Hello,
1) for proxies, the recommendation is 1 tasks per core. for repositories, the recommendation is 3 tasks per CPU core. That value was increased recently to reflect capabilities of modern CPUs (just in case you had lower numbers in mind). Keep in mind, that you also need resources for the operating system.

2) As you already have "tons of proxies" with "one per host", I would use them. If resources are not enough, you can still decide whether you to add more resources. I would keep it simple for small environments (yours still sound small to me) and just make it one all-in-one server. I don't like the complexity of distributed components.

4) I'm a "direct storage access" fan. "better" depends on the situation... if you struggle with VM-stun, then BfSS is the way to go. If not, direct NFS can be even faster overall, because there are no storage snapshots involved (which take some seconds). For HotAdd with NFS, you really need one proxy per host. But I would never do HotAdd with NFS anyway.

5) it depends on the customer situation, but in general yes. yes, only one network card - I don't like the complexity managing two network interfaces.

Best regards,
Hannes

[matteu]

Thanks for your answer.

1) My issue is :

Lot of hardware have big CPU for repository but disk array can't handle this. How do you "manage" it ?

2) ok

4) Thanks !

5) Thanks perfect

[HannesK]

1) well, if disks are too slow, I see two options. Use faster disks or live with it I'm not sure, how a software should fix the ratio between compute and IO performance.

I assume that you are using REFS / XFS if you are using backup modes that merge or create synthetic full backups.

[matteu]

Sorry but I need some details :p

For my customer, I will use direct NFS.

I'm thinking about configure my proxy with 2 vNIC. 1 with backup vlan and 1 with storage VLAN. Is it "mandatory" or just optional ?
On the documentation it's not written "mandatory"
https://helpcenter.veeam.com/docs/backu ... ml?ver=110
"The backup proxy used for VM data processing must have access to the NFS datastores where VM disks are located. "

I could stay with 1 vNIC if both VLAN are routed or performance will not be good ?

[tdewin]

In general you want to avoid routing for your backup traffic. It's not a support issue from our side but backup streams are one of the most bandwidth consuming workloads in your DC so you probably don't want to have them pass through your whole network stack impacting everything else.

[HannesK]

I like to avoid dual-NIC configurations. Too much work for little benefit. The user guide is correct: "access" is the key word. One NIC is possible, so yes, you "could"

Proper datacenter switches do routing at line-speed (I never worked with SMB switches, probably they are slower). My main concern would be that it sounds like you still plan virtual proxies. So you go up and down all the hypervisor stack... I prefer to avoid that complexity.

In IT I only know "too slow" and "fast enough". I cannot estimate "good performance" via a forum

[matteu]

Yes sorry, I forget to give some description.

VBR server : 1 virtual (backup vlan)
proxy : 10 virtual (backup vlan)
Repository 2 physical (backup vlan) 1 for backup and the other for backup copy
Storage NFS (storage vlan)

This is the architecture.

What should I do for the best performance ?

I can route storage + backup vlan or add 2nd card on proxy for storage vlan

It's arround 300 VMs on 10 ESX host

If routing is not good, I suppose the best is to add a second vnic

[HannesK]

whether routing is good or bad depends on the hardware. the easiest way it do just do it and see whether it's "fast enough"

[matteu]

Thanks for your answer
I will try with 2 vnic and I will see it !

Now they are arround 150MB/s and they have arround 300VMs. I know the speed will be really better, I will see how much on the first try next week!