Comprehensive data protection for all workloads
Post Reply
jkivits
Lurker
Posts: 1
Liked: never
Joined: Nov 27, 2014 9:33 am
Contact:

ReFS Repository Sizing

Post by jkivits »

Hi all,

I am looking for some advice sizing hardware for a new windows-based repository. I haven't found relevant whitepapers or topics on these forums yet, so I hope you guys can help me. I have been looking at hardware from Dell, HP and Cisco and while they all look like they could suffice, I do have many questions regarding the disks, and I hope anyone here can help me based on their experience. Basically, how many disks do I need and what size?
First off, our environment:
• About 40TB of data (VMware)
• 10% change rate per day
• I need a primary retention of 14 days, secondary depends on whether Tape will be implemented for offline retention
• We have 2 datacenters, so we will go for 2 repositories
• On 1 of our 2 datacenters I would like the ability to copy the backup to tape, so in the future there needs to be the possibility of connecting a library to this server
• We have 10Gb networking

Our current repositories are dedupe appliances, which are far too slow for a descent restore, which is why we are looking for windows repositories. Without dedupe we need a many disks for the capacity, but probably as well for the throughput.

Q1: With ReFS, is reverse incremental the way to go? Or forever incremental? How many synthetic fulls should I make, once a week, or daily?
Q2: How much data savings is normal for ReFS in a datastore? Can I just assume one full and 13 incrementals?
Based on the Dell R740xd2, which can house 24 LFF disks, I would like to use 16 slots (so we can expand later on if needed) of 8TB. That give 128TB Raw, minus parity etc. Disks may be larger if the answer to Q2 is given.
Q3: Should I put those disks in a hardware raid, or Windows Storage space? Will raid6 give enough performance to saturate 20Gb/s network? (with 20Gb/s I can do a full backup/restore in 5 hours
Q4: Will there be enough IOPS for surebackup and instant-on restores?
If more IOPS are needed, the UCS might be a better solution as it can handle more disks
And last topic: Encryption. We want our data to be encrypted. Veeam has a function to encrypt the data, but so does windows, and so do the disks (Self Encrypting Disks). My guess is that Veeam encryption is the best option for us, as we most likely will use Tape as well (and therefor the data remains encrypted).
Q5: Is my guess right?

In case I have missed out anything else worth looking into, I'm open for suggestions.
nmdange
Veteran
Posts: 527
Liked: 142 times
Joined: Aug 20, 2015 9:30 pm
Contact:

Re: ReFS Repository Sizing

Post by nmdange » 1 person likes this post

Q1 - forward incremental is recommended over reverse. Using forever forward is the best option unless you have a large number of restore points. Forever forward has a single full backup and the oldest incremental is merged each day. I use forever forward with 14-day backups and weekly synthetic full for 60-day backups. I don't think you would ever want to create a synthetic full more than once a week.
Q2 - Space savings mainly come from when you have multiple synthetic fulls, essentially they will take up as much space as incrementals would otherwise. The bigger benefit to ReFS is speeding up synthetic operations because data does not have to be rewritten.
Q3 - I have found good success using RAID 60 with a hardware raid controller with battery-backed write cache, with 10 drives per RAID 6. I would generally not use storage spaces unless you are going to use storage spaces direct with SSD/NVMe caching (but then you need 4 servers and 10gbps networking may not be fast enough)
Q4 - this really depends on workload and is hard to say. Keep in mind you can attach a JBOD to the Dell server to get more drives. Also I would highly recommend looking at Supermicro for your storage hardware. That is what I use and it costs significantly less than the other vendors. I have a 36-bay 4U Supermicro server and a 44-bay JBOD for 80+ drives total, although I have a much larger dataset than you do.
Q5 - Veeam encryption will ensure the backups stay encrypted even if the backup files are copied to different media manually.
DaveWatkins
Veteran
Posts: 370
Liked: 97 times
Joined: Dec 13, 2015 11:33 pm
Contact:

Re: ReFS Repository Sizing

Post by DaveWatkins » 1 person likes this post

If you're worried about IOP's and disk space Dell do the MD3460 and MD3860 chassis that are 4U and can hold 60 drives. Pair it with a 1U server with enough speed and you should be set, and if you need more you can daisy chain expansion chassis that hold another 60 drives
Post Reply

Who is online

Users browsing this forum: itmedia, Semrush [Bot] and 113 guests