Netbackup SLP Performance

[wrgomes1991]

Hello all,

I'm wondering if someone could help or already had this scenario.

I have a netbackup media server connected to a Data Domain and a Tape Library so I created a SLP who backup the image on disk (DataDomain) and duplicate to the tape with a bigger retention the issue was the performance of this duplications they took a long time to duplicate from DataDomain to the tape.

Someone experienced the same or have any suggestion to get a better performance ?

Regards

[X2]

@wrgomes1991SLPs are merely workflow control feature, with different options e.g. backup, duplication, retention, replication, etc. that can be configured during the lifecycle. The performance of writing to tape should be independent of you using an SLP or a traditional policy.

Have you tried to manually do a duplication from the DD to the tape? How does that peform?

[wrgomes1991]

Thanks for your Reply !

I already did a manually duplication and the same poor performance.

I will try to do a backup direct to the tape and monitor the performance maybe the proccess to take the image from the DD to the tape was more slowness for some reason.

I don't have much expertise at how the DD works as well.

[Mouse]

SLP itself is just a controller. You need to look into your data path and find out the bottleneck. Typical issues when duplicating from Data Domain:

• Using shared 1GbE interface between Data Domain and the Media Server. For many modern fast drives the performance from this combination may not be sufficient if the actual throughput falls below drive's minumum write speed so it will have to resort to shoe-shining of tape (google this for more details if not familiar). I recommend to use iperf to check Data Domain to Media Server throughput, it's built into DD OS and available for all major operating systems, see https://support.emc.com/kb/334570 for detailed instructions
• Buffer tuning values are not optimal, check https://www.veritas.com/support/en_US/article.100000498.html for Unix and https://www.veritas.com/support/en_US/article.100030830.html for Windows
• DD OS version and DD Boost plug-in version may be not up to date and/or best performing configuration, check what is the latest recommended versions and upgrade if required
• Check if the media server that writes tape is the same media server that reads from the Data Domain. Sounds obvious but more frequently than not I find SLPs where Alternate Read Server is not set so a random media server picked for reading the image and another media server for writing, sometimes the bandwidth is limited to 1Gbps or, worse, link passes a firewall and this results in poor performance
• There are recommendations in the Dell EMC Boost integration guide around configuring media servers, check them out

[Genericus]

I have been doing this for years, and it works excellently.

I was writing to two DD990 and from there to 18 LTO5 tapes via FC. You have to be carefull due to HBA limitations, we ended up with two 990 so we could spread the load. I could (mostly) drive the LTO5 at speed, although we could definitely overload them.

I now have a DD9800, and I can simultaneously backup two Oracle DB (over 40TB), into the data domain, and write out to tape at the same time, my last tape completes duplicating within 30 minutes of the backup completing.

Seeing the backups writing to the data domain at over 200MB/Sec per channel still cracks me up, it is ludicrous speed, and it writes out to tape as fast as the tape can go.

You might want to maximize the memory and cpu on your data domain, and if it is running it's cleaning cycle that can also impact performance. 

[Nicolai]

Make sure you use a new version of Data Domain OS. 

Previous on DD990, re-hydration performance would degraded with age. So a brand new backup would stage to tape at a acceptable speed, but a e.g 6 month backup would stage to tape with no more than 40-60 MB/sec. This had to do with fragmentation of unique data block in the data domain. Newer versions of DD OS may have fixed this issue. 

As a work around - stage long term backup to tape right away, and accept you will have two copies of that data, for a shorter or longer period of time.