VBO architecture needed for 1000 mailboxes and 30TB of O365 data

[squebel]

Hello,

Where can I find some sure-fire recommendations on what would be required, from a VBO architecture perspective, to backup 1000 mailboxes and 30TB of data?

[flavio.santos]

Hello squebel,

I was using the Veeam calculator: https://calculator.veeam.com/vbo

You connect with your customer credentials and you will have a recommendation.

[squebel]

Thanks for pointing this out flavio.

How do we get around the throttling that Microsoft has in place for these calls so we can pull down these backups as quickly as possible?

[Mike Resseler]

@squebel

There is a procedure to request (temporarily) to reduce the throttling. (https://community.veeam.com/blogs-and-p ... ttling-720)

Also, have a look here: https://bp.veeam.com/vbo for more advice on the architecture

[JRRW]

I can't stress this enough: Keep on Microsoft to make sure they don't throttle you. I had to beat them over the head nearly weekly for a bit with a hickory stick to keep them from throttling. One support rep said they removed it, then the next, then it was for sure removed and then it came back the next week, and then finally I got it escalated to someone who got it right. Luckily, VBO's logs call it out pretty well if throttling is occurring.

Also, make sure you have enough CPU assigned - my Mailbox proxy+repository has 6 cores of a Xeon Gold 5122 and my OneDrive/Sharepoint proxy has another 6 of the same proc - with Hyper-threading that leaves me with 4 threads available for host and a 2-vcpu object storage repository.

On that note: If you do local (Block not object) I highly recommend trying to break it up some vs doing all 30TB into one repository. It CAN work, but depending on your retention, local / block is using jet database(s) and can be very slow. I wouldn't do anything less than a raid-50 personally if it's NL-SAS, maybe raid-10 if you can afford it (but personally I like the raid-50 for the faster reads) - or just go all SSD (which really isn't that expensive for 30TB if you do 7.68TB Enterprise SATA SSD in a Raid-5 w/hot spare). I also run my C drives for the two VMs on a raid-1 SSD so that the proxy/app/logging is not impacting the database performance at all. That may be overkill but I like to think it helps.

Otherwise if you can get it or have it already, object storage does way better and gets better storage usage as well.

One other thing I did to keep resources a bit 'cleaner' for scaling, is used a dedicated VM on a different system to handle management of the jobs - that way it never steals resources from the 'workers' - and my repo/Proxies are all Server Core (all the management/log collecting/updating can be done remotely from the management server so i rarely have a reason to log into them)

[Mike Resseler]

Thank you Ryan for your great advice and write-up here!