Some Guidance on Best Practise

[scott_mac]

Hi All,

Hopefully some more experienced users can offer me a little guidance. I'm relatively new to the world of Virtual Machines and the backup thereof. We're using VMware Essentials and after a botched attempt to use Backup Exec i've seen the light and have purchased Veeam Essentials.

Basically we have the following environment:

- 2 x ESX hosts
- 5 x VM's hosted on them, 3 x Server 2008 R2 (~300GB each), 1 x Server 2003 SBS (1TB) and a Linux one (2.5GB) for vSphere Mobile Access from my iPad.
- Datastores are all on a Netgear ReadyNAS 3200 in iSCSI mode.
- Onsite backups are to another ReadyNAS (Pro 4) over a LAN
- Offsite backups are to be replicated using Netgear's Replicate application for the ReadyNAS products (this appears to be a shiny front end to rsync!). This will be over a WAN that has a 100Mb circuit for upload and around 20Mb download at the other end due to infrastructure limitations beyond our control.

I've been using Veeam for around 6 weeks for onsite backups only that are then periodically copied to tape for offsite storage - however the time taken to do the tape element lead me to look at the replication of the ReadyNAS environment.

I've read through the FAQ but am still not 100% certain what the best way for me to go is - the main concern being the changes and the replication between the ReadyNAS devices.

Firstly, deduplication is a puzzling one, depending on what you read it seems to suggest grouping all the servers into one job will offer the best dedupe, but at the cost of speed... presently I have one job per server, with a time stagger so there is only ever one running at a time - I opted for this as I felt it was the best precaution for the possibility of the backup failing half way through due to, say, a power cut, which would be bad!

Secondly, so far i've been using Incremental backups with a Periodic Full every month which has been fine for onsite (though strangely despite the retention and deleted being left at the default 14 days I have a lot more than 14 VIB files...) However now that I need to replicate over a WAN i need to be wary of actual file sizes. The Incremental would be fine, except for the new 'Full' that will appear every month as the largest server has over 1Tb in it which leads to a 500Gb VBK file - copying this over the WAN is basically impossible. Would a synthetic full (something else i'm not quite sure on) combined with rsync be a viable way to go, or...

Given that the Replicate from Netgear uses rsync, i'm thinking i might be better off with Reverse Incremental of all servers in one job, but if i'm honest don't really understand the benefits of each - despite reading as much as I can on it!

Can anyone help!?

Thanks

Scott

[scott_mac]

Although reading further on Reverse Incrementals, there seems to be a risk with those also, in that if one backup fails or gets interrupted than the main VBK file will also be corrupt and thus you may not be able to restore from it.

:S

[pcrebe]

Hi scott_mac,
off topic but important: is one of yours 5 x VM's a Secondary Domain Controller? If is so your are ok for restore SBS2003.

Bye,
Carlo

[scott_mac]

Hi Carlo,

No, our secondary DC is a physical box on the LAN - just in case! As said, i'm a little new to the Virtual world and figured this was safer - it was already in place before we virtualised so kept it.

Thanks

Scott

[pcrebe]

Hi Scott,
perfect! because veeam perform an non-authoritative restore of AD (as you probably know) and if you have only 1 DC (often in SBS lan) is really difficult restore AD, anyway, in the future, migrate the DC2 to a local storage of one host backing it up with veeam. IMHO

For your first question i can't help you because i don't use off site backup for the moment.

Bye,
Carlo

[scott_mac]

Thank you for the advice anyway Carlo - that is useful to know!

I am hopeful someone can answer my main question

Scott

[Vitaliy S.]

Firstly, deduplication is a puzzling one, depending on what you read it seems to suggest grouping all the servers into one job will offer the best dedupe, but at the cost of speed... presently I have one job per server, with a time stagger so there is only ever one running at a time - I opted for this as I felt it was the best precaution for the possibility of the backup failing half way through due to, say, a power cut, which would be bad!

Provided that there are no jobs running concurrently, there is no win in timing at all.

If you want to optimize the backup window you can run 3-4 backup jobs concurrently (depending on the hardware of your backup server). However be aware that the more jobs you start, the more CPU load you will have, so make sure that CPU usage is not maxed out, otherwise the performance bottleneck would be in the backup server itself.

In all other cases, it is highly recommended to group your VMs for better dedupe rates.

Secondly, so far i've been using Incremental backups with a Periodic Full every month which has been fine for onsite (though strangely despite the retention and deleted being left at the default 14 days I have a lot more than 14 VIB files...)

That's more than expected behavior. Please look through this thread for more info: Rollback points and full backups

Would a synthetic full (something else i'm not quite sure on) combined with rsync be a viable way to go, or...

Definitely.

Given that the Replicate from Netgear uses rsync, i'm thinking i might be better off with Reverse Incremental of all servers in one job, but if i'm honest don't really understand the benefits of each - despite reading as much as I can on it!

If you're using rsync as a replication engine and cannot afford sending entire VBK file over the WAN link, then I would definitely suggest choosing reversed incremental backup mode, as it only requires updating a single VBK file.

By the way, take a look at this topic in case you haven't seen it already, pretty useful: incremental backups or reverse incremental??

Although reading further on Reverse Incrementals, there seems to be a risk with those also, in that if one backup fails or gets interrupted than the main VBK file will also be corrupt and thus you may not be able to restore from it.

That's not correct. It does not "corrupt" the main VBK, you can still use it. Here is an existing topic with the same question:
Reverse incremntal - if failed - how is it handled?

Hope this helps!

[scott_mac]

Vitally - thank you very much. I had tried searching but as I didn't really know what I was trying to search for I didn't get too far!

Plenty of reading for me there and it seems that Reverse Incrementals are the order of the day.

I read on one thread here that if you run all of the VM's in one job it does still treat them as separate backups, so if it failed for some reason part way through the job, any complete servers that had already been done would be fine, it would only be the ones during and after that would require using the previous backup - so one job with all servers in is recommended, with reverse incremental.... though my vCenter server will have to be done differently as I have to connect directly to the host given it is virtualised!?

I shall read the links you provided first but I believe that is the way to go!

Thanks again

Scott