How best to use a DR4100 / Physical vs Virtual Repos?

[KickedAbyss]

My environment:

• 4-VM Hosts (1tb total RAM, 20c/host)
• 170tb Compellent SC8000 3-Tier array
• Dell DR4100 (not currently used for Veeam)
• Veeam B&R Enterprise

Currently I have three backup jobs:

• Job 1 - 43VMs at 12tb total -- daily Forward Incremental w/weekly Synthetics
• Job 2 - 22 VMs at 19tb total (mainly databases) -- 6-hour Reverse Incremental
• Job 3 - 5 VMS at 1tb total -- daily Forward Incremental

All use High compression and WAN Target for storage optimzation to get as much dedupe as possible; my OS break down is about 90% 2012r2, 10% 2008r2 so I realize some substantial dedupe savings per job.

We moved from vRanger to Veeam in a bit of a rush, so our vRanger data is/was on a DR4100 (36tb raw), so I have this setup for Veeam:

• Host 1 - VeeamCore Server/Repository-1 (for Job 2 and 3) that also acts as a Proxy for that host
• Host 2 - Proxy and Repository-2 (for job 1 data) - two different servers
• Host 3 - Proxy
• Host 4 - Proxy

The Repositories are using RDM to my spindle tier on my Compellent - the 1st and 2nd are all SSD - and all servers are running 2016 Datacenter, utilizing ReFS for the repository drives. Full 2x10gb networking per host for data, and 2x10gb for iscsi. 80gb to Top of Rack (backup vlan is separate from host IPs so I THINK it goes to ToR during backups, not positive)

Performance thus far isn't bad, my Veeam systems just use a fair amount of CPU/RAM on my virtual hosts; each Proxy is 6c/5gb RAM and the Repo/Proxy/Veeam core is 8c/22gb, with another 4c/6gb to the 2nd Repo. I run 2016 Datacenter-Core for all but the main veeam server which helps a ton not having a full OS installed (but I'll give vRanger props for having virtual appliance proxies that can be provisioned from the vRanger client to your hosts)

Storage wise I'm not using all that much; with my current retention set of around three months, I'm actually only using 20tb for all 3 jobs, as ReFS and 9.5 is amaaaazing.

We've hit retention limit with the DR4100 so I can repurpose it fully and wipe it for Veeam, instead of hammering my SAN and iscsi network by running backups from/to itself (we're very well aware it's not best practice, but we had plenty of storage on the SC8000)

My question(s) is around best practices for larger environments on physical vs virtual, and whether it's best to dedicate a host to Veeam. My thoughts are these:

• Keep current Veeam architecture but move the Repositories onto CIFS share(s) on the DR4100 and leverage its dedupe/compression & Replication (I have an identical DR4100)
• Keep current Veeam Architecture and install 2016 on DR4100 hardware and use DAS or Storage Spaces to provide backup repository.
• Repurpose DR4100 hardware as a 5th host in my cluster and run a single Repo/Proxy off it, offloading all compute/storage from my production environment and leveraging the DAS 36tb - the DR4100 has Xeon procs and something like 48gb RAM, as it's basically an R720dx with a custom OS installation

I've gone over the user guides but there's very little information out there on sizing Veeam and what these best practices might be.

I'd love the communities input on this, or even pointing me at any documentation I have't seen that should answer this. I've only seen a few people use the DR4100's and there isn't much in terms of what i might expect on dedupe/compression using veeam. For vRanger and their RDS format, I was seeing ingest rates of 1GB/s and savings of 13x (90%) which was ridiculous. Unfortunately, vRanger itself is junk, and the DR4100 / vRanger combo had nearly constant issues with properly cleaning up dedupe data and applying retention policies, as vRanger actually used the DR4100 OS to handle the file management. It was horrible, to be honest. WHEN it worked, it was very nice. But it broke nearly every 2-3 months in some fashion.

[Andreas Neufert]

The DR4100 is a dedup device and by design of those we can not process synthetic operations. (no protocoll like DDBoost or Catalyst supported).
That means if this will be your first backup target you need to run active full (read 100% data weekly) + incrementals with it.

I don´t know if it would be a good idea to install Win2016 on it. It depends on if this is supported and if there is a good Raid Controller (battery buffered). If there isn´t look if the Storage Controller is on Micrsofts HCL to run something like Storage Spaces on it.

I would do the following.
Use the existing Server to backup.
Place the DR4100 on a second fire section and add a good LAN connection to it.
Use our Backup copy jobs to backup data to the DR4100. You can use the same settings/Setup as described here: https://www.veeam.com/wp-quantum-dxi-be ... veeam.html

[Andreas Neufert]

Oh and you can achive the nearly the same savings on the DS4100 if you turn of all optimizations/datareduction on the Veeam side and let the DR4100 do everything. But that is not the idea. The idea is to have fast and reliable backups and when you can reduce the data amount before you transport the data to the DS4100... the overall data held on it would be the same, the inteface would show a not so good reduction ratio on the DR4100, but performance would be better.

[tinto1970]

I don´t know if it would be a good idea to install Win2016 on it.l

the DR4100 (i have worked with it in my previous company) is simply a PowerEdge server with a few of not as fast hard disks. Its price is really high because of the software (formerly Ocarina Software) installed on it.

My results were really poor. Sure you can spend half (or even less) of the money and pick up a normal PowerEdge with more disks/more space/more speed, install windows 2016 and enjoy the feautures of ReFS, which is widely more modern, performant and reliable than the DR4100 dedupe's software.

[KickedAbyss]

It's full 10Gb to the DR4100, so it sounds like my better option is to reprovision it. Yes, it has an H710p or similar on it, as well as a cache card (not sure how that would be leveraged by Windows). It's as Tinto said, basically an R720dx. Has 10k raid-1 in the back to run the OS off of, too.

The hardware is basically 12x1tb drives in the main unit, and a DAS Powervault running another 12x3tb. I can either then use storage spaces to handle the drives, or just present a virtual disk via the Raid Controller. Either way, i'll get good performance (I'm not sure if storage spaces has any advantage as I won't have SSD/15k to 'tier' it off) using ReFS on it.

So the follow up question would be how best to leverage it in terms of a Veeam solution. If I move the primary Veeam operational side to it, and run it as the main repo, I could in theory just have proxy servers on the current VMWare and let them just toss data off onto the Repo over LACP/Bonded 2x10gb connections. That should give fast backup, and give a solid offload of i/o on my SAN. I'd lose the dedupe/compression but honestly the unit (DR4100 not their newer one) was really struggling with the amount of ingest we were doing with vRanger, so i'd rather let Veeam handle the dedupe and compression side. I have Enterprise licensing, so I can then setup my DR4100 in my DR site with Veeam and just let it replicate the system straight across.

Does the repository server handle any of that dedupe/compression or is it all on the proxy servers?

And one question for another forum: Can you utilize the same repository for both CloudConnect AND BD&R?

Thanks for everyone's input.

[Andreas Neufert]

Dedup and Compression is mainly done by the Proxy. Part of the dedup engine runs on the Repository. Both "inline".

CC and B&R can share a Repository Server but not the exact same folder. You have to create 2 folders and create for each a repository role.
Important the B&R Servers are not aware of any jobs that run against the Repository from each other. Take care that you have enough resources (RAM mainly).
As the components on that system can run only one version, both B&R need same version.

I would say that it is best practices to keep the CloudConnect Hosting seperated from the B&R Servers that takes care of other stuff at the hosters.

[veremin]

And one question for another forum: Can you utilize the same repository for both CloudConnect AND BD&R?

I would advice against such design. Both backup servers must run the same version, otherwise component version mismatch will certainly occur. And with SPs having to leverage always the latest version and being provided with updates sooner than ordinary customers do, this task becomes trickier. Thanks.

[xarses]

Can you use w2016 with refs for synthetic backups and then move this to dr4100 with a backup copy job?
We have a dr4100 and the performance with synth backups are poor....

[Andreas Neufert]

Hi you can backup to a no dedup repository and use Backup Copy to to the DR4100 without active full processing by enabling the feature within backup copy job: "Read the entire restore point from source backup instead of synthesizing it from increment" This feature avoid any kind of syntetic processing on the target.