Slow backup to tape (LTO6)

[stevendt]

Hi,
I am new to Veeam B&R. Using Version 10 in a home setting, so looking to use the Community Edition.

I have managed to get hold of a second hand Dell LT2000 Autoloader with an IBM LTO6 Drive.

I have got Veeam B&R working and using the tape drive and library, but writing to tape is much slower than I had expected. Using File to tape to backup a Network share, I am getting about 60MB/s. That can't be right surely?

I initially used the Native drivers in Windows Server 2016 but have also tried the Vendor Drivers (For Windows Server 2012R2) which made no difference.
I have also tried turning off the "Use hardware compression when available" option
The Tape Drive and Tape Server is on the same machine as the main B&R application.
I am backing up the shares over a 1GB link, but the Bottleneck is reported as the target.
The data is Windows Server Backup and Novastore Backup files - not sure if they are already compressed or not.

Can anyone help me improve the tape writing speed please?
regards
Dave

[stevendt]

mmm, maybe this is as just about as good as it can be on a 1GB LAN ???

If all of the (lightly loaded) network bandwidth was available to the backup job . . .
1Gigabit/s would roughly be 100,000 bytes/second (I usually use a rule of thumb of 10 bit/byte allowing for overhead)
100,000B/s would be (ignoring that 1024 thing) 100MB/s maximum
I would have hoped for more, but may not actually be able to get much over the 60MB/s that I am seeing?

I would be interested in comments, but I will try a local drive to tape transfer and see what I get

regards
Dave

[soncscy]

Hi Dave,

What does the bottleneck say? If it's network, then likely you get the idea Is there any ways to either improve that or just connect the tape hardware to the share hardware?

[stevendt]

Hi, thanks a lot for the reply.

As I mentioned, the bottleneck is reported as “Target”, i.e., the tape drive.

Unfortunately, I can’t connect the tape hardware directly to the NAS. It’s a cheap consumer device with no SAS ports.

Oddly, earlier today, I did a file to tape transfer from a local SAS drive and it was even slower at about 40MB/s, but that was a relatively small data set of about 60GB. The local SAS drive was a 10k RAID1 array through an HP ProLiant controller, I was really surprised how slow it was?

Regards
Dave

[gurneetech]

Have you tested max speed read from the network share?

How is the Library connected, is it sharing a SCSI bus with other hardware that also has significant activity?

Otherwise: here are some comparisons.
We're an HP shop. We have dedicated physical backup servers with disk storage and 2xLTO6 tape libraries directly connected.
SAS6 connected tape libraries, 2x on single HBA. HBA has 4 lanes, so in our case each library has full bandwidth available
ReFS datastore (this matters as it can impact read rate)
All backups follow the D2D2T model.
Running VBR 9.5U4
We have several different tape out jobs.
One job which I would call optimal because of the dataset usually reports a processing rate of ~130MB/s and a throughput (all time) of ~140MB/s
Jobs otherwise report a processing rate around 90MB/s and throughput rates near the 160MB/s theoretical max.
Example of Load: Source 67% > Proxy 2% > Network 44% > Target 60%

[Dima P.]

Please check the network connection between Tape Server and source server. When you are running file to tape jobs trafic goes from the source over the network to the Veeam Tape server component (the one that's set in the Veeam B&R console for your tape device) and only after if goes from tape server to the tape device. Cheers!

[stevendt]

Hi Gurneetech, thanks for the reply and your metrics.

I will try to find a mechanism for testing the speed of the network share.

The Library is connected through a 6Gbs SAS interface, it's the only device on the link. In fact, the SAS board and the Library both have dual ports and it was originally connected (not by me) with cables in both ports. This meant that I saw two devices for the Tape and Library in Device Manager which was confusing. I don't think that the dual link is meant for throughput, so I have disconnected the second link. (It made no difference to the speed anyway. It's a Dell TL2000 Library and the documentation suggests that only 1 link is needed between the SAS card and the library.)



Dima, thanks for input. The connections between all of the device are fine - Gigabit Ethernet with no error showing on the switch statistics. The data from the source server goes through a couple of switches to the Veeam Server which hosts the Tape Server and is directly connected to the library

regards
Dave

[Dima P.]

Dave,

Can you run a test file to tape job directly from Tape Server to compare the throughput? Cheers!