When running backup job from NetApp filer CIFS shares, job data rate is very slow when compared with other jobs not using CIFS.
Example speeds from NetApp: 352MB/min, 427MB/min, 597MB/min, etc
Example speeds from other NAS/workstation: 2,512MB/min, 5,785MB/min, 2,641MB/min, etc
Server is up-to-date; Backup Exec 16, FP2, HF128051 running on Windows Server 2016 Standard.
Anti-virus has the following exceptions: C:\Program Files\Veritas\Backup Exec
Server has 4x Gigabit NICs teamed to provide 4Gbps throughput and 3Gbps SAS connection to LTO-7 autoloader.
LAN speed test to NAS with 10MB test file: write 529MB/min and read 1.84GB/min.
This is not the first time this has occurred and there has been no change of hardware or network elements.
Software encryption and software compression is enabled for both scheduled job and test job.
One-time backup from CIFS to disk storage yields similar job rate; 167MB/min.
We have a duplicate setup at our other site where speeds from the NetApp to BE16 seem to be much better but the only real difference is the actual files being backup up are different the the autoloader is LTO-6
It is expected to see differences in performance for CIFS vs NDMP - in fact performacne of CIFS backups is why we added support for NDMP backups. Basically the CIFS protocol was created for individual file access over shares and not really designed to stream large amounts of data in one go, the NDMP protocol was designed with backup devices and moving streams of data in a way that is optimized for such devices, in fact our remote agent also uses NDMP (but of course is not running on the disk storage device where the shares are)
CIFS (Share level) Backup
- Pros: files backed up up in same way as any files on local disk so can be redirect restored more easily
- Cons: incrementals and differentials may not work as expected and performance is compromised
- Pros: Speed of backup (especially if tape library attached to NMDP device) and a type of differential or incremental sequence can be run, but it does work in an odd way
- Cons: restores must be done to same type and version of NDMP Environment (OS on the array)
If you are comparing CIFS with a location where our remote agent runs, then same info applies
If you are compareing CIFS on one vendor's device with another vendor's device then that is down to the hardware and not Backup Exec
If you are compareing CIFS with an iSCSI connection to a NAS (actually making it into a SAN connection) then again we would expect to see differences in performance
You can test/benchmark read from the CIFS share using tar for example or any other tool that you like.
The simplest way is to use MobaXterm and run :
dd cf /dev/null myfs
and look for windows network usage to know the real time speed or this could be easily encapsulted in a script :)