Multiple VMWare proxies & Backing up SQL servers

[aaron@ARB]

Hi all,

Just testing Veeam at the moment (from Backup Exec, so forgive me) and after having ran a full backup on the weekend of 25 vms I have some questions that I was hoping someone could possibly answer for me.

My environment consists of 3 Dell Equallogic members which suffice to say are decidedly rapid, a Dell R720xd which has a 10gb iSCSI connection direct to the SAN and R10 local storage which I am using as the backup repository.

1) At the moment I have just the default installed backup proxy which is the local machine. The backup server I am using does not struggle with load when running backups with utilisation about 30% when backing up. Throughput into the repository is about between 6-10gb/min give or take. Given I don't think anything is struggling, would there likely be any performance gain if I used another R720 that I had bought but not actually used yet.. Its specifications are similar to the R720XD in that it has 2x6 core Xeon some-a-rathers, 16gb RAM and a 10gb connection direct to the SAN. If I use this server, can i possibly increase the ingestion speed of data into the backup repository? Or would it only be of use if the bottleneck was the source?

2) in our environment we have a SQL server that runs our transactional ERP system. When I back it up with backup exec, given its over 1.5tb in size, by the time the actual backup has completed, the size of the snapshot is quite large even when being backed up in a low time (we are a 24 hour company) and the machine would 'stun' for up to 2-3 minutes at the final stage of the snapshot deleting process while it wrote the redo redo log (so to speak). To get around this issue, we would run agent based (so frontend) backup which would not use VMWare snapshots and the issue went away (at the expense of backup speed). Given that there are no agents with Veeam for such purposes, what is the suggested way to backup an SQL Server that seems to generate a lot of snapshot data in an effort to not 'stun' the server?

3) Wtih regards to dedupe and compression (all options selected are just defaults), is there any advantage in having one job with 25 VM's listed vs 25 jobs with 1 vm listed? I recall being told that dedupe is more efficient if all the VM's are listed in one job, is that the case?

4) Given the size of my backups and the data involved, I have had the ability in Backup Exec words to have the full backup starting at 16:00 Friday afternoon and then have all the incrementals running from 21:30 Monday -> Thursday. Is there a way in veem to have different start times for a full vs incremental backup? At the moment I manually adjust the start time.

Cheers.

[Shestakov]

Hi,
1) You are correct, for the performance gain, it`s better to investigate the bottleneck statistics first. Could you provide yours?
Using Direct SAN backup method and highly productive proxies can speed up the backup process.

2) Here on the forums are a several related topics about SQL Server Backup Best Practices. You can start with this one.
If your storage is high-powered in terms of IOPs, you will not have issues with snapshots.

3) Correct, combining VMs running same OS and applications is recommended for the best dedupe ratio.

4) You can choose any time and date to start the job, as well as an option to start the job before/after another one; but, if you want your job to start at different time during the week, you need to use a powershell script.

Thanks!

[aaron@ARB]

Hello again,

1) Going through the last full backup and the incrementals that I am running now, they some examples are below
18/04/2015 7:00:49 PM :: Load: Source 66% > Proxy 53% > Network 48% > Target 7%
18/04/2015 5:50:35 PM :: Busy: Source 51% > Proxy 57% > Network 63% > Target 0%
18/04/2015 6:48:14 AM :: Busy: Source 84% > Proxy 60% > Network 63% > Target 29%
18/04/2015 4:23:28 AM :: Busy: Source 97% > Proxy 38% > Network 2% > Target 0%
18/04/2015 5:34:52 AM :: Busy: Source 95% > Proxy 42% > Network 10% > Target 6%
18/04/2015 3:34:01 PM :: Busy: Source 89% > Proxy 50% > Network 15% > Target 0%

given that the source is showing as the main reason for the slowdown, would another well spec'd proxy actually help? or is the issue going to be getting the data out of the SAN quickly? what is the above showing me?

2) I will have a read. The SAN is actually dedicated for this Db and application. Its good for about 4000 IOPS but its still going to be an issue when the snapshot redo log is written and the VM is stunned. I really want to backup the server without an image level backup i think. is something like this possible?

3) Okay, i will leave them combined. Is there a whitepaper on how that actually works? How does it make a difference if they are together or not given that since the jobs are limited by the number of active processes the proxy can handle, not all the VM's are being backed up at once so how does it make a difference?

4) no problem, i will look at this, its not a hardship to change manually but if i did want to do it via powershell, i am assuming there is a guide somewhere? Also..perhaps a feature request? To be able to schedule full and incremental jobs independently of each other.

[Shestakov]

1) These numbers show percent of time the given data processing stage was busy versus waiting for other stages to provide or accept the data. Do not expect the numbers across all processing stages to add up to 100%, as busy time of each processing stage is measured separately and independently.
Looks like your main bottleneck is the "Source" - production storage disk reader component. The percent busy number for this component indicates percent of time that the source disk reader spent reading the data from the storage. For example, 97% busy means that the disk reader spent almost all of the time reading the data, because the following stages are always ready to accept more data for processing.

It looks like you've reached the max performance of your source datastore. You can also try to use hotadd proxy, compare the numbers and then stick to transport mode with best performance rates. What are the processing rates for those jobs?

3) VMs of same OS with similar applications have a lot of similar data blocks, which are deduplicated by Veeam Backup Proxy. Basics are covered in the related User Guide section. More info can be found on the forum.

4) Powershell scripts are discussed in the related subforum. There are lots of handy scripts, take a look! Here is a one we are talking about.
And yes, we will take it into consideration as the feature request. Thanks for the feedback!