network traffic using Direct SAN

[Dave338]

Hello, I am configuring a new veeam BR v8 server, and I see the same behavior. Previously I had been using several versions of Backup Exec, so I'm new to Veeam world.
This is my config:

Physical server HP Gen8 with 4 core CPU, and 12GB RAM, with dual HBA connected to the SAN (EVA P6550) at 4Gbps
Windows 2012 R2 with dedup enabled on datastore.
I use Direct SAn access, and have disabled failover to network, so direct SAN should be the only option.
VMWare datastores are presented to the veeam server and I see them on disk admin, but I have them "offline" so win cannot touch any data, is this correct??

When backing up, I see no ethernet traffic on windows task manager, but if I go to the Resource Monitor, I see network traffic over 1gbps towards local machine (server name shown). I don't know how to verify that is using DSAN mode, I don't see it at job statiscs... only see that at every server log for virtual disks... is that correct???

27/01/2015 13:07:38 :: Using backup proxy VMware Backup Proxy for disk Hard disk 1 [san]

Anyone has an idea if this is normal or not???

[Dave338]

Hello again.

It seems nobody is using Direct SAN or this behavior is not very common

I've been testing with the new performance parameter in v8-patch1, that one

Code: Select all

Performance enhancements
Added experimental support for direct data movers communication when both are running on the same server (for example, when backing up to a local storage on backup proxy server). If your local backup jobs report Network as the bottleneck, and you see high load on some backup proxy server NICs when the data was supposed to stay local to the server, your jobs performance is likely to benefit from this behavior modifier. To enable alternative data exchange path, create the DataMoverLocalFastPath (DWORD) registry value under HKLM\SOFTWARE\Veeam\Veeam Backup and Replication, and set it to the following values:
0: Default behavior (no optimizations)
1: Data exchange through TCP socket on the loopback interface (faster)
2: Data exchange through shared memory (fastest)

All jobs are executed with datastores presented at host, and seen offline at windows disk manager. No disk or network load is seen at ESX in any case.
Here you have what I see with datamover value = 0 (default), I first run the same job to have the data on controller's cache, these are runs from 2 to 4 . First run took 3:58 minutes

Run 2, the job took 3:17 at 109MB/s, you can see the high traffic in network, although veeam says is using san mode.


Run 3 with datamover=1, same behaviour, job took the exact same 3:17 at 110MB/s so speed is virtually identical


Run 4 with datamover=2. And with this option, NO network traffic is shown at any level, so it seems to work and data bypass network. The job took 3:15 at 120MB/s, so a little speed up but could be caused by workload on the SAN


Sooooooo with this testing, it seems that is normal to see that traffic at network, but seems is internal traffic, because windows shows nothing at inbound/outbound ethernet.

Could someone confirm that??? Is sure to go with datamover=2 at this point?? (i feel more comfortable seeing no network traffic xD )

Regards.
Dave.

[cpeheotel]

If it says your traffic is using [san] then that indicates Veeam is pulling bytes and bits direct from your SAN luns. If it is pulling through the VMware management network stack you'll see [nbdssl] Also, the disks should be 'online' in disk management. Just don't initialize them. If your SAN lets you, present them as Read Only to the backup server for extra safety.
-CCP

[alanbolte]

Hello Dave,

It looks like your backups are running with the proxy and repository roles assigned to the same server. These roles always involve running at least two separate VeeamAgent.exe processes for each running backup job, even on the same server. VeeamAgent.exe is the "data mover" described in the patch notes. In earlier versions of the product, these data movers worked exactly the same regardless of whether they were running on the same or different servers, so they communicated with each other over the network stack. We've been progressively adding optimizations for the special case of two data movers on the same server, but this registry value (when set to 2) is the first change to move communication completely off of the network stack. That said, the default behavior shouldn't actually cause traffic to hit your switch, but it can register on your NIC because the server is processing it through network loopback.

[Dave338]

Hello.

Thanks for the replies.
Then is exactly what I thought, because no traffic is seen at windows network, but at performance monitor shows movement.

I'll stay with value 2 for the moment, so is working very well.

One doubt... cpeheotel said that I should put 'online' the disks in windows disk manager, but I've read surfing on the web that with windows 2008R2 and above (mine 2012R2), the default SAN policy is offline shared, so is good to stay offline and LUN presentation "standard" (read/write), or better online and LUN presentation read only?? I think it should be the same, my traffic is going throug SAN, so my config should be fine I guess..

Regards.
Dave.

[foggy]

No need to bring disks online, Veeam B&R requires read-only access to the LUN, which is provided when it is in offline mode. However online disks are also ok as soon as they are not initialized, since it is disk initialization that can result in data corruption. And since Veeam B&R automatically sets SAN Policy to Offline/disables disk automount during installation of proxy server, this prevents disks from being initialized.