Direct Restore to Microsoft Azure | VeeamPN software-defined networking
Post Reply
randerson999
Influencer
Posts: 14
Liked: never
Joined: Jun 30, 2016 11:49 am
Full Name: Ross Anderson
Contact:

VCCe VM Size on Azure (for Merge Performance)

Post by randerson999 » Sep 10, 2018 1:18 pm

I'm still struggling to find the recommended size for the Veeam Cloud Connect VM when loaded on Azure. I've been running a POC for a month or two using Standard A2 (2 vcpus, 3.5 GB memory) and about 9TB of disk space (3 x 3TB HDDs striped with ReFS) but the merges are taking 6-7hrs with only 30-40GB of nightly changes (~1TB total backup size).

This seems way slow to me so I'm wondering if just the VM itself is undersized .. i was under the impression that using ReFS would cut WAY down on merge times, but in this case it has not. Or is it just the fact that I'm using HDD attached to the VCC VM that is causing the slow merge.

Any thoughts, recommendations or documentation that you can point me to please? (BTW - yes I've read many, many setup docs - but can't find a good sizing best practice).

Thanks!

randerson999
Influencer
Posts: 14
Liked: never
Joined: Jun 30, 2016 11:49 am
Full Name: Ross Anderson
Contact:

Re: VCCe VM Size on Azure (for Merge Performance)

Post by randerson999 » Sep 11, 2018 2:31 pm

Addendum: I'm looking at the following Azure options for the VCC VM. What am I missing here and why doesn't Veeam have a specific recommendation? (the docs only say that 8gb RAM is recommended and I believe i read 4 vCPUs somewhere). I'm happy to spend more if I'm actually getting something for that extra money, but I don't want to pay for higher end CPUs if we're never using them or need them. I think storage throughput is probably the deciding factor on the VCC VM?

A4_v2 - good price, doesn't support premium disks if we need more IOPs or expansion, no local SSDs, available IOPs based on number of drives (500 per)
B4ms - great price, seems like a good option since backups only happen sparingly during the day, supports premium disks
D4s_v3 - slightly higher cost, more ram, higher supported disk throughput

ejenner
Expert
Posts: 403
Liked: 60 times
Joined: Mar 23, 2018 4:43 pm
Full Name: EJ
Location: London
Contact:

Re: VCCe VM Size on Azure (for Merge Performance)

Post by ejenner » Sep 11, 2018 3:33 pm

Not sure whether or not Veeam have a size-guide for Azure but can't you find what is holding it back by looking at the logs?

randerson999
Influencer
Posts: 14
Liked: never
Joined: Jun 30, 2016 11:49 am
Full Name: Ross Anderson
Contact:

Re: VCCe VM Size on Azure (for Merge Performance)

Post by randerson999 » Sep 11, 2018 4:05 pm

Good question - unfortunately I can't look back through the logs again because our POC subscription has ended already - so I need to create the environment again from scratch. When I inspected the VCC VM during a merge, I could not see a bottleneck on the CPU so I have to assume it was storage related. Cost-wise the B series (B4ms_ is the cheapest but I'm wondering if having a burstable VM will create future problems. On the other end, the D4s_v3 looks to have the performance we need but at extra cost - plus I don't know that we really need the extra CPU performance (hence, the higher cost). This is the quandary.

anthonyspiteri79
Veeam Software
Posts: 610
Liked: 145 times
Joined: Jan 14, 2016 6:48 am
Full Name: Anthony Spiteri
Location: Perth, Australia
Contact:

Re: VCCe VM Size on Azure (for Merge Performance)

Post by anthonyspiteri79 » Sep 17, 2018 12:11 am

Though the forums you will find threads talking about the viability of Azure for medium to large VCC repositories. General consensus is that you will run into performance issues for heavier IO operations such as merges. There are a lot of factors that go into this, but disk performance/contention is going to be the biggest bottleneck.

With the introduction of the Archive Tier in Update 4 for Backup & Replication 9.5 the idea is that the requirement for large footprints for shorter term retention will drop and the bulk of storage will be tied to Azure Blob. You don't get around the IO issue in terms of merges, however a smaller footprint means less cost for the repository meaning you could think about throwing higher performance disk at the Azure instance.
Anthony Spiteri
Global Technologist, Product Strategy | VMware vExpert
Email: anthony.spiteri@veeam.com | Mobile: +61488335699
Twitter: @anthonyspiteri | Skype: anthony_spiteri

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest