Instant recovery performance boost guide

[xh63]

Hello all

We are redesigning our backup solution to incorporate immutable backup solutions in our infrastructure to fight back the ransomware attack.
When I testing the instant recovery, I was surprised (in a good way)by the performance, the vm is up and running in less than 1 min with high performance( can't see any latency).
Well done Veeam 11
Then I started reading, found the following which may provide the answer. The below is for veeam v10.
"Through the updated Veeam documentation and material, we examined, for the Instant VM Recovery, what the keys are, in making the operations faster, and cost-effective. Veeam put the focus in three areas. The first one was the increasing RAM cache size to 1GB. Moreover, this RAM cache was optimized to measure all of the read blocks by flushing out the blocks that have been read least often, reducing latency, and enabling faster performance on recovery. Another focused area is the intelligent prefetch of data blocks. "

Now I need to get a quote for the hardware and needs to get some suggestions or answers for the points below
1. Can someone supply kb that have much more details on how exactly veeam v11 improve the IR performance
2. How to design a solution to improve the IR performance economically
3. The Memory sizing guide for the backup server to improve the IR performance. If I want to run 30 vms in IR, how much ram do I need
4. Whats the requirement for the backup repo. RAM, disk, caching. We are using centos 8 as the immutable repo. will lvm caching help the IR performance?

Many thanks in advance

David

[HannesK]

Hello,

1. the main performance boost comes from V10. In V11 we only did stability improvements and the normal "maintenance tasks" that are common in software development. The What's new document of V10 and V11 (helpcenter.veeam.com) list all that information. The changed default compression algorithm in V11 might also help.
2. Use a server with internal disks / SSDs.
3. Nobody can predict how much IOPS 30 VMs need. I would start with 64 GB of RAM and go to 128 GB RAM if needed.
4. Every caching helps. Whether it's from the RAID controller or from the volume manager. I'm not an LVM expert as I always recommend RAID controllers with proper amount of cache & BBU. That setup is proven to work.

Best regards,
Hannes

[Andreas Neufert]

I think one of the main performance boosters for Instant VM Recovery that is environmental, is storage/network latency.
For example when you boot your servers from Linux Repository to a Windows vPOWER NFS (mount server) and then go by NFS to the ESXi hosts, you have 2x network latency in there.

Make sure you have at least 10GbE in there. Work with you network team to optimize as much as you can network latency. Firmware Updates, Driver Updates, Windows Updates, Avoiding Firewall/Routers in between,...

[xh63]

Hi Hannes
Thanks for the reply
1. the main performance boost comes from V10. In V11 we only did stability improvements and the normal "maintenance tasks" that are common in software development. The What's new document of V10 and V11 (helpcenter.veeam.com) list all that information. The changed default compression algorithm in V11 might also help.
I went to helpcenter, but then which one should I choose to get to the v10 whats new.
2. Use a server with internal disks / SSDs.
Is this your opinion or veeam recommendation? Can you be more specific? I can't give that kind of suggestion to my business.
3. Nobody can predict how much IOPS 30 VMs need. I would start with 64 GB of RAM and go to 128 GB RAM if needed.
In my post, I referred to a paragraph that says "what the keys are, in making the operations faster, and cost-effective. Veeam put the focus in three areas. The first one was the increasing RAM cache size to 1GB." Where do they get that information from? increasing ram cache per running IR vm? so need 30GB for 30 vms? what is it?
4. Every caching helps. Whether it's from the RAID controller or from the volume manager. I'm not an LVM expert as I always recommend RAID controllers with proper amount of cache & BBU. That setup is proven to work.
If I have to design a solution rely on the caching on the RAID controller, this would be the end of world for us. how to scale up for the least?

Andreas, thanks for the input.

[HannesK]

Hello,

1. yep

2. I'm not sure how to be more specific without knowing your workloads. 30 high transactional SQL servers are different then 30 file or print-servers. Feel free to double check the "Veeam" best practices at https://bp.veeam.com/vbr - but it's the same thing: it's impossible to answer that question with that amount of information. And everybody will say something else depending on the personal price / value "feeling".

3. You can test by changing the IRReadCachePerDiskMB (DWORD) in HKEY_LOCAL_MACHINE\SOFTWARE\Veeam\Veeam Backup and Replication reg key. But I'm relatively sure that it will only help very little if you increase it.

4. sorry, not sure what you mean.

As a general advice: please don't consider instant recovery as a general purpose disaster recovery solution. We offer Veeam Replication or CDP for that.

Best regards,
Hannes

[xh63]

could you please read the question 1 and 4 more carefully and reply accordingly?

[xh63]

put this in a simple way, your suggestion no.4 is a very bad one. if you can understand this.