Primary backup on dedup WITHOUT landing-zone - really?

[mdiver]

I often see customers being advocated by specific hardware vendors to use dedup appliances for every backup need - including primary backups.
As also a trainer I try recommending the usage of dedups without a landing-zone for secondary backup only.
In the new slide deck for VMCE2020 there is also a slide saying exactly that. This is in sync with my experiences on e.g. recovery speeds.

I would like to gather some experiences from the community on that. Especially regarding DD with boost integration.
Anyone happy with this configuration for primary? What are your recovery speeds (full-VM/single objects). What caveats did you meet?

Thanks,
Mike

[HannesK]

Hello,
short answer: no.

Inline dedupe is only for backup copies. Nothing has changed in the reference architecture. https://www.veeam.com/blog/what-is-the- ... cture.html

I know very few happy customers. But only those who almost never do restores and if they to restores, then they only restore very little data. The majority has "limited happiness". Restore speed and merge speed is not good.

Best regards,
Hannes

[mdiver]

Thanks. This is fully in sync with my experiences.
I was just unsure I missed something because it gets so heavily pushed by the vendors.

Best regards,
Mike

[HannesK]

sure - they want to sell

I would even argue, whether dedupe appliances pay off for secondary target vs. REFS / XFS...

post368298.html

15.4 TB used on disk
99 TB size on disk
6.4 data reduction factor based on block cloning
12.8 data reduction factor if to compare with dedupe boxes where we store data uncompressed (2x on average)

So we can get 1:13 ratio for free with REFS or XFS.

[Regnor]

I can support your opinion; never do primary backup on a dedup appliance.

@Hannes:
For long term retention (~2 Years+) I still would trust a dedup appliance more then a REFS storage in regards of stability.
And they could offer a better protection against ransomware if you use a different protocol (DDBoost, Catalyst).

[softflame]

According to Veeam, any Veeam Ready - Repository has been tested and met restore performance requirements.

For example Quantum's DXi meets the restore performance requirements, and also supports Fast Clone for faster synthetic full creation. Should be fine as a primary backup target.

Protecting against ransomware is a broader topic, but suffice to say just relying on an obscure protocol to protect you is pretty scary.

[sys-adm]

We have since 2 years ExaGrid as primary backup storage and last year we switched also our second backup appliance. ExaGrid has per design a landing-zone and it works great.
Before we used ExaGrid our backup appliance was a HPE StoreOnce and this device is horrible.
Never buy a HPE backup device. Slow and we losed all 6 months our backup chain.

[datadrop]

From my experience over the years, I do not like dedup appliances as a primary target. As a secondary target yes, but as a primary target - definitely no. Just consider running a Sure Backup Job running on a dedup appliance, a device which surely doesn't cope well with random IO. Back in roundabout 2006, when dedup appliances came up - these systems were planned to replace tape. One of the first systems available were for example SEPATON systems ("No Tapes" spelled backwards). So, when I set up backup with a dedup target, I try to keep random IO (writing or reading) off these targets.

[SalSolo]

I agree with the technical assessment and recommendations .. however as many things in life this is not black and white.
I see several customers choosing to use dedup appliances as primary targets because of budget restrictions .. they cannot afford both a premium landing target and a dedup appliance and often choose to go with the largest capacity as it is more important for them to protect large amount of data than quick restore. This goes for many of the small companies as they are encountering that same demands from the business as the large ones today as data is key for all companies no matter the size ..but small companies tend to not have a solid IT budget yet so they are forced to make some compromises. It is easy to recommend the best if there is no pricetag on it.

I think the most important part here is to educate the customer and the sales people to ensure that they understand the tradeoff and those that do understand are generally happy with dedup appliances. In my experience those with the bad experience is because of bad saleswork so the expectations has been set wrong .. and yes DD boost is one of the ones I see misused a lot. It is a great feature but like everything else it it not something that just solves all issues and work in all circumstances.

[Regnor]

According to Veeam, any Veeam Ready - Repository has been tested and met restore performance requirements.

For example Quantum's DXi meets the restore performance requirements, and also supports Fast Clone for faster synthetic full creation. Should be fine as a primary backup target.

Protecting against ransomware is a broader topic, but suffice to say just relying on an obscure protocol to protect you is pretty scary.

I can't say much about Quantum as I don't know their solutions.

Regarding ransomware protection; it's only a part of a solution as everything else is.

[ChrisSnell]

ExaGrid is, of course, not a dedupe target. ExaGrid is tiered backup storage, so the Landing Zone gives customers the fast backup target without dedupe (and providing fast restores), then customers benefit on cost from having dedupe storage in the retention zone. All inside a single appliance that does all the moving about work for them, no need for messy copy jobs.

Plus the Retention Zone is immutable, and within a few months ExaGrid will have a retention time-lock too. So the immutable functionality prevents data being changed or over-written, and the time-lock prevents data being deleted. Veeam customers almost always use the DataMover integration, providing a secure Veeam share, not CIFS or NFS - much more secure.

Give ExaGrid a test next time you're looking at backup storage for large Veeam environments.

[miketate]

We are currently in the market for a Data Domain and have been engaged with Dell who have encouraged that we use the data domain as the 'primary' backup destination. We have read through the Veeam best practices and whilst we note that Veeam does state to use a 'non-dedupe' device as the primary, there is also contradictory info in article https://www.veeam.com/kb1745 i.e. under the section;

"Special Scenarios Based on Vendor Own Guidelines + Backup Job With Active Full or Synthetic Full Backup: The above vendor best practices and design guides mention situations where a deduplication storage can be used as primary backup target." - our backup method would be utilizing orever incrementals with periodic synthetic fulls.

There seems to be some concerns around 'DD Boost' not being as great as it's advertised, what particular issues or concerns do you have with this?

[mitchellm3]

I'll bite...Our organization uses StoreOnce for Primary, Secondary and Tertiary copies of backups. We backup and copy 4900+ VMs in an 8 hour window. The backup and replication speeds are not an issue. Obviously our situation is different than others. We started out with FC storage as our landing spot many years ago. However, space was the issue. Once we moved to a metro cluster solution for primary storage, instant VM recovery went down on the priority list. That allowed us to transition to a dedupe appliance for primary and secondary backup storage. As we started to use HPE StoreOnce and ransomware entered the discussion, the security benefits of the catalyst protocol started to shine.

In our organization we don't do restores every day. Most restores our users need they can do themselves since they are file share based...whether that be restoring from previous versions in teams/onedrive or restoring from previous version on our NetApp file shares. When it comes to our application teams, we have a process where we do VM snapshots before upgrades so that provides the faster restore option if something goes wrong. But, if we have to restore a VM, the restore speeds from a dedupe appliance is acceptable.

So in the event of a disaster of a data center, we would much rather bring up the VM in a crash consistent state in our secondary DC than do an instant VM recovery of an app consistent VM that is missing upwards of 24 hours of data. For the systems that are critical to our organization that are highly transactional, they all have their own built in mechanisms to handle outages with log shipping and things of that nature.

To answer the question on speeds, we just did a restore from a storeonce across the WAN and got 135MB/s. The VM had a single 90GB disk. So 15 minutes to restore isn't bad. I do understand the concerns about restoring a single VM with a 5T disk but that is a risk we accept. To be honest, that is also a limitation with VEEAM in that they are single threaded per disk. We do have Commvault set up to use storeonce as well. We are testing catalyst but our current setup uses NFS. Commvault is able to use multiple threads and can restore at 568MB/s (4.5Gbps).

So if we didn't have a metrocluster solution where we had VMs in both data centers and we were not able to protect all the VMs with some other storage replication technology (yes I know that is not a backup) I would change out the primary backup target to something that could handle the IVR as that would be a big part of the DR strategy.

[Ayhan-Co]

Hello All,

What are the effects of using Data Domain as Primary Backup Repository (Short Term & Long Term retention)? Does anyone configure like this?

I know the deduplication effect of DD to backup & restore jobs. What are other effects? I wonder that because Veeam Best practice guide use DD as long term backups.

Regards,