- Posts: 24
- Liked: never
- Joined: Nov 17, 2020 9:49 pm
- Full Name: Martin B
We have defined a Backup JOB (Active Full + incrementals) for some VMs wunning in VMWARE Hosts. Some VMs are small 200GB (application, etc) and other ones are big 2TB (file server, FTP Servers, etc).
All in one JOB.
Our VEEAM Server (v10) acts as repository, and has a 100TB volume (NTFS) in a dedicated SAN Storage (ME4012 12x12TB NL-SAS RAID6), set to process per-VM backup chains, 8 tasks. We have also 4 HotAdd proxy, defined for 4 tasks each.
The JOB invokes the 4 proxys via 10G.
First thing to note: I´ve seen that task limit value (in the repository), define the amount of VMDKs all proxys can process (per-vm or not set), is it ok ?
The Active FULL JOB start very fast, for at least 20/30 minutes (1.7GBs or 1.4GBs), but after the small VMs are backed up, the JOB drops abruptly to lower speeds, while backing up file servers, Proxys CPU utilization very low, and network utilization very low, at some point storage performance reported by the DELL console 240MBs, windows resource monitor reports 100% utilization over the repository drive.
It seems that when VEEAM asigns a big VMDK to a proxy, start processing at a X speed, but as its big it takes time to process, meanwhile other small VMDKs are processed, but when they get done, speed of the big VMDK not increases....
But If I define a specific JOB to only backup for example 2 File Servers in one, speeds are higher.
Could it be the tasks ratio defined ?
Proxys 4 each (6 vCpu each), Repository 8 (has 16 cores).
Should I reduce the amount of proxys ? tasks ?
Could it has to do with mixing small and big VMs in one JOB ? Its better to take big VMs in a separate JOB ? But how can I control Per-VM amount of streams ?
Could it be a saturation somewhere ?
I´ve tried DiskSpd writing 6 streams and it gives 6 x 200MB/s (Storage max is 1200MB/s in any given combination of streams)
Is there a way to tell VEEAM the max Per-VM chains writting at any given time ? For example to restrict to not more than 4 VM at a time. Something similar to the repository tasks for VMDKs, but for Per-VM chains.
If someone has a clue or some experience with such behavior will help me a lot !!