Scaling veeam

[stevenj]

While I recognise this is a how long is a piece of string type Q.....has anyone got some approximate rules of thumb to give me an idea how many vmware guests per veeam 4cpu appliance? say 40 guests per appliance per 10 hour backup window?

regards

[Alexey D.]

There is no limitation on how many VMs you can process with your B&R installation. Concerning backup window: it depends on your backup windows as well as actual environment, specifically VMs' size and content, amount of daily changes in virtual disks, production storage, backup target storage, job processing mode, etc.

Basically you should simply test backup performance on subset of your environment and go from there.

[Vitaliy S.]

Just wanted to add, provided that you cannot fit into your backup window with a single installation, you may install a second Veeam backup server and manage both serves from a single console - Enterprise Manager. Thanks!

[Gostev]

My rule of thumb is 5TB of source VM data per Veeam Backup v5 server on 8 hours backup window with vSphere ESX 4.x, assuming mixed VM workload of nearly 100% virtualized shop with decent hardware.

Start planning your Veeam Backup deployment around these numbers, make a pilot deployment, then adjust the number either way depending on pilot deployment results. I've seen up to 15TB of mixed workload VMs processed by a single backup server with direct SAN access mode (fast SAN and good backup server hardware). Can even do significantly more that that, but with specific workload only (for example processing very large, fairly static file servers).

[stevenj]

While I recognise this is a how long is a piece of string type Q.....has anyone got some approximate rules of thumb to give me an idea how many vmware guests per veeam 4cpu appliance? say 40 guests per appliance? per 10 hour Replication window?

regards

[Gostev]

To put it short, Veeam server is never a bottleneck in case of replication. WAN link speed always is. And in case you are replicating over 1Gb and to full ESX, then what I said above in this thread fully applies to replication as well.

[stevenj]

Thanks


However we are using ESXi....so the blockage/bottleneck is ESXi? cant get past 50MB/sec per ESXi host...and the WAN isnt 25% full....

regards

[Gostev]

Сorrect.

[stevenj]

I am just running a test to ESX and ESXi.....I am finding there is no difference in performance, 3MB/sec is it....seems awfully sklow for a LAN connection.

Is there anything I can check to look for an [setup] error?

Looking at real time stats and html report they are both identical in throughput and show no errors.


regards

[Gostev]

Try uploading a file to ESXi with vSphere Client - Datastore Browser and see what speed do you get.

[stevenj]

Full backups seems slow, I guess its because Veeam is telling us how fast its processing the image as opposed to data transfer.....Also I see a 80% difference between Linux guests and Windows guests....3 v 5 MB/sec for fulls.

My second incremental pass showed 301MB/sec on ESX and 88 on ESXi for Linux guests.

[Gostev]

Yep, 3 times difference between ESX and ESXi, exactly as expected.

Don't rely on processing rate to compare throughtput between VMs, it is no good for this - affected by too many other things. As for full backup throughtput when moving virtual disk data, it depends on VMDK content (specifically, amount of empty blocks) and not guest OS.

[ctchang]

Just wanted to add, provided that you cannot fit into your backup window with a single installation, you may install a second Veeam backup server and manage both serves from a single console - Enterprise Manager. Thanks!

Just a quick related question:

Say I have 4 sockets licenses, the above means I can setup (install) multiple Veeam V5 backup server as much as I want to load balance the loading without license being counted right? CPU sockets license only courts towards ESX host, but not Veeam backup server right?

Also, I can connect these multiple Veeam backup server to the same MSSQL database right?

Thanks,

[ctchang]

Yep, 3 times difference between ESX and ESXi, exactly as expected.

Don't rely on processing rate to compare throughtput between VMs, it is no good for this - affected by too many other things. As for full backup throughtput when moving virtual disk data, it depends on VMDK content (specifically, amount of empty blocks) and not guest OS.

Anton,

or the old question is when can we have a real processing speed report showing the real speed of the backup windows?

[Vitaliy S.]

Say I have 4 sockets licenses, the above means I can setup (install) multiple Veeam V5 backup server as much as I want to load balance the loading without license being counted right? CPU sockets license only courts towards ESX host, but not Veeam backup server right?

Yes, that's correct.

Also, I can connect these multiple Veeam backup server to the same MSSQL database right?

You can use the same SQL Server, but you cannot share the same Veeam configuration database.

[ctchang]

Thank you very much.

[stevenj]

Scaling veeam......um....

Looks like to scale Veeam the max I can do in a 10 hour backup window (6pm to 4 am) is 25 guests per quad core virtual server (assuming 2 replications).........this means to do 150 vmware guests I need 6 x quad core machines at the source end (which will be at 100% CPU across all 4 CPUs), that is a huge impact/cost on the VMware infrastructure, is anyone else seeing this sort of scaling issue?

The other way is to go physical at source with SAN connect.....so if I have a 2 x hex core server with 3.4ghz CPus will veeam scale across all 12 cores? I might get away with 1 maybe 2.....

....then I dont know what I need at the DR end..........I suspect I need several (2 or more) dedicated small ESX servers installed just to do the de-compression....single quad core socket and 4gb in multiples.....that's not cheap either.......

Is this what others are finding?